Journal of Financial Economics 108 (2013) 1–28

Contents lists available at SciVerse ScienceDirect

Journal of Financial Economics

journal homepage: www.elsevier.com/locate/jfec

The other side of value: The gross profitability premium$

Robert Novy-Marx
Simon Graduate School of Business, University of Rochester, 500 Joseph C. Wilson Blvd., Box 270100, Rochester, NY 14627, United States

a r t i c l e i n f o abstract

Article history: Profitability, measured by gross profits-to-assets, has roughly the same power as book-
Received 15 December 2011 to-market predicting the cross section of average returns. Profitable firms generate
Received in revised form significantly higher returns than unprofitable firms, despite having significantly higher
29 May 2012
valuation ratios. Controlling for profitability also dramatically increases the performance
Accepted 6 June 2012
Available online 26 January 2013
of value strategies, especially among the largest, most liquid stocks. These results are
difficult to reconcile with popular explanations of the value premium, as profitable firms
JEL classification: are less prone to distress, have longer cash flow durations, and have lower levels of
G12 operating leverage. Controlling for gross profitability explains most earnings related
anomalies and a wide range of seemingly unrelated profitable trading strategies.
Keywords:
& 2013 Elsevier B.V. All rights reserved.
Profitability
Value premium
Factor models
Asset pricing
Quality investing

1. Introduction for market cap and B/M, there is a positive relation

Profitability, as measured by the ratio of a firm’s gross
profits (revenues minus cost of goods sold) to its assets,
has roughly the same power as book-to-market (B/M)
predicting the cross section of average returns. Gross
profits-to-assets is also complimentary to book-to-mar-
ket, contributing economically significant information
above that contained in valuations, even among the
largest, most liquid stocks. These conclusions differ from
those of earlier studies. For example, while Fama and
French (2006) finds that earnings has explanatory power
in Fama and MacBeth (1973) cross section regressions,
Fama and French (2008, p. 1663) finds that ‘‘profitability
sorts produce the weakest average hedge portfolio
returns’’ among the strategies they consider and ‘‘do not
provide much basis for the conclusion that, with controls
$
I thank Gene Fama, Andrea Frazzini, Toby Moskowitz, the
anonymous referee, and participants of the National Bureau of Economic
Research Asset Pricing and Q Group meetings.
E-mail address: robert.novy-marx@simon.rochester.edu
between average returns and profitability.’’ Gross profit-
ability has far more power than earnings, however, pre-
dicting the cross section of returns.
Strategies based on gross profitability generate value-
like average excess returns, even though they are growth
strategies that provide an excellent hedge for value.
The two strategies share much in common philosophi-
cally, despite being highly dissimilar in both character-
istics and covariances. While traditional value strategies
finance the acquisition of inexpensive assets by selling
expensive assets, profitability strategies exploit a different
dimension of value, financing the acquisition of produc-
tive assets by selling unproductive assets. Because the
two effects are closely related, it is useful to analyze
profitability in the context of value.
Value strategies hold firms with inexpensive assets
and short firms with expensive assets. When a firm’s
market value is low relative to its book value, then a stock
purchaser acquires a relatively large quantity of book
assets for each dollar spent on the firm. When a firm’s
market price is high relative to its book value the opposite

0304-405X/$ - see front matter & 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jfineco.2013.01.003
2 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28
is true. Value strategies were first advocated by Graham
and Dodd (1934), and their profitability has been shown
countless times since.
Previous work argues that the profitability of value
strategies is mechanical. Firms for which investors require
high rates of return (i.e., risky firms) are priced lower and,
consequently, have higher book-to-markets than firms for
which investors require lower returns. Because valuation
ratios help identify variation in expected returns, with
higher book-to-markets indicating higher required rates,
value firms generate higher average returns than growth
firms (Ball, 1978; Berk, 1995). While this argument is
consistent with risk-based pricing, it works just as well if
variation in expected returns is driven by behavioral
forces. Lakonishok, Shleifer, and Vishny (1994) argue that
low book-to-market stocks are on average overpriced,
while the opposite is true for high book-to-market stocks,
and that buying value stocks and selling growth stocks
represents a crude but effective method for exploiting
misvaluations in the cross section.
Similar arguments suggest that firms with productive
assets should yield higher average returns than firms with
unproductive assets. Productive firms that investors
demand high average returns to hold should be priced
similarly to less productive firms for which investors
demand lower returns. Variation in productivity in this
way helps identify variation in investors’ required rates of
return. Because productivity helps identify this variation,
with higher profitability indicating higher required rates,
profitable firms generate higher average returns than
unprofitable firms. Again, the argument is consistent with,
but not predicated on, rational pricing.
Consistent with these predictions, portfolios sorted on
gross profits-to-assets exhibit large variation in average
returns, especially in sorts that control for book-to-
market. More profitable firms earn significantly higher
average returns than unprofitable firms. They do so
despite having, on average, lower book-to-markets and
higher market capitalizations. Because strategies based on
profitability are growth strategies, they provide an excel-
lent hedge for value strategies and, thus, dramatically
improve a value investor’s investment opportunity set. In
fact, the profitability strategy, despite generating signifi-
cant returns on its own, provides insurance for value.
Adding profitability on top of a value strategy reduces the
strategy’s overall volatility, despite doubling its exposure
to risky assets. A value investor can, thus, capture the
gross profitability premium without exposing herself to
any additional risk.
Profitability also underlies most earnings related
anomalies, as well as a large number of seemingly
unrelated anomalies. Many well known profitable trading
strategies are just different expressions of three basic
underlying anomalies, mixed in various proportions and
dressed up in different guises. A four-factor model,
employing the market and industry-adjusted value,
momentum, and gross profitability factors, performs
remarkably well pricing a wide range of anomalies,
including (but not limited to) strategies based on return
on equity, market power, default risk, net stock issuance,
and organizational capital.
The remainder of the paper is organized as follows.
Section 2 provides a simple theoretical framework for the
prediction that gross profitability predicts the cross sec-
tion of expected returns and shows that the predicted
relation is strong in the data. Section 3 investigates the
relation between profitability and value more closely. It
shows that controlling for book-to-market significantly
improves the performance of profitability strategies and
that controlling for gross profits-to-assets significantly
improves the performance of value strategies. Section 4
considers the performance of a four-factor model that
employs the market and industry-adjusted value,
momentum, and gross profitability factors, and it shows
that this model performs better than standard models
pricing a wide array of anomalies. Section 5 concludes.
2. Profitability and the cross section of expected returns
Fama and French (2006) illustrate the intuition that
book-to-market and profitability are both positively
related to expected returns using the dividend discount
model in conjunction with clean surplus accounting. In
the dividend discount model, a stock’s price equals the
present value of its expected dividends. Under clean
surplus accounting the change in book equity equals
retained earnings. Together these imply the market value
of equity (cum dividend) is
X1
Et ½Y t þ t dBt þ t 
Mt ¼ , ð1Þ
t¼0
ð1þ rÞt
where Yt is the time-t earnings, dBt  Bt Bt1 is the
change in book equity, and r is the required rate of return
on expected dividends. Holding all else equal, higher
valuations imply lower expected returns, and higher
expected earnings imply higher expected returns. That
is, value firms should outperform growth firms, and
profitable firms should outperform unprofitable firms.
Fama and French (2006) test the expected relation
between profitability and expected return with mixed
results. Their cross sectional regressions suggest that
earnings are related to average returns in the manner
predicted, but their portfolio tests suggest that profit-
ability adds little or nothing to the prediction of returns
provided by size and book-to-market.
Fama and French (2006) employ current earnings as a
simple proxy for future profitability, however, and gross
profitability is a better proxy. Earnings in Eq. (1) represent
a firm’s true economic profitability. Earnings off the
income statement represent a firm’s true economic profit-
ability reduced by any investments that are treated as
expenses, such as research and development (R&D),
advertising, or human capital development. Expensed
investments directly reduce earnings without increasing
book equity but are nevertheless associated with higher
future economic profits and, therefore, higher future
dividends. When considering changes to earnings in
Eq. (1), it thus makes no sense to hold all else equal.
Gross profits is the cleanest accounting measure of
true economic profitability. The farther down the income
statement one goes, the more polluted profitability
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28
measures become, and the less related they are to true
economic profitability. For example, a firm that has both
lower production costs and higher sales than its compe-
titors is unambiguously more profitable. Even so, it can
easily have lower earnings than its competitors. If the firm
is quickly increasing its sales though aggressive advertis-
ing or commissions to its sales force, these actions can,
even if optimal, reduce its bottom line income below that
of its less profitable competitors. Similarly, if the firm
spends on research and development to further increase
its production advantage or invests in organizational
capital that helps it maintain its competitive advantage,
these actions result in lower current earnings. Moreover,
capital expenditures that directly increase the scale of the
firm’s operations further reduce its free cash flows rela-
tive to its competitors. These facts suggest constructing
the empirical proxy for productivity using gross profits.1
Scaling by a book-based measure, instead of a market-
based measure, avoids hopelessly conflating the produc-
tivity proxy with book-to-market. I scale gross profits by
book assets, not book equity, because gross profits are an
asset level measure of earnings. They are not reduced by
interest payments and are, thus, independent of leverage.
Determining the best measure of economic productiv-
ity is, however, ultimately an empirical question. Popular
media are preoccupied with earnings, the variable on
which Wall Street analysts’ forecasts focus. Financial
economists are generally more concerned with free cash
flows, the present discounted value of which should
determine a firm’s value. I, therefore, also consider profit-
ability measures constructed using earnings and free
cash flows.
2.1. Fama and MacBeth regressions
Table 1 shows results of Fama and MacBeth regres-
sions of firms’ returns on gross profits-to-assets, earnings-
to-book equity, and free cash flow-to-book equity.
Regressions include controls for book-to-market (log(B/M)),
size (log(ME)), and past performance measured at horizons
of one month ðr 1,0 Þ and 12 to two months ðr 12,2 Þ.2 Time
1
Several studies have found a role for individual components of the
difference between gross profits and earnings. For example, Sloan (1996)
and Chan, Lakonishok, and Sougiannis (2006) find that accruals predict
returns, and Chan, Lakonishok, and Sougiannis (2001) argue that R&D
and advertising expenditures have power in the cross section.
Lakonishok, Shleifer, and Vishny (1994) also find that strategies formed
on the basis of cash flow, defined as earnings plus depreciation, are more
profitable than those formed on the basis of earnings alone.
2
Gross profits and earnings before extraordinary items are Compu-
stat data items GP and IB, respectively. For free cash flow, I employ net
income plus depreciation and amortization minus changes in working
capital minus capital expenditures (NI þ DP  WCAPCH  CAPX). Gross
profits is also defined as total revenue (REVT) minus cost of goods sold
(COGS), where COGS represents all expenses directly related to produc-
tion, including the cost of materials and direct labor, amortization of
software and capital with a useful life of less than two years, license fees,
lease payments, maintenance and repairs, taxes other than income taxes,
and expenses related to distribution and warehousing, heat, lights, and
power. Book-to-market is book equity scaled by market equity, where
market equity is lagged six months to avoid taking unintentional
positions in momentum. Book equity is shareholder equity, plus deferred
taxes, minus preferred stock, when available. For the components of
3
series averages of the cross sectional Spearman rank correla-
tions between these independent variables are provided in
Appendix A.1 and show that gross profitability is negatively
correlated with book-to-market, with a magnitude similar to
the negative correlation observed between book-to-market
and size. I use Compustat data starting in 1962, the year
of the American Stock Exchange (Amex) inclusion, and
employ accounting data for a given fiscal year starting
at the end of June of the following calendar year. Asset
pricing tests consequently cover July 1963 through Decem-
ber 2010. The sample excludes financial firms [i.e., those
with a one-digit standard industrial classification (SIC) code
of six], though retaining financials has little impact on the
results. Independent variables are trimmed at the 1% and
99% levels. The table also shows results employing gross
profits-to-assets, earnings-to-book equity, and free cash
flow-to-book equity demeaned by industry, where the
industries are the Fama and French (1997) 49 industry
portfolios.
The first specification of Panel A shows that gross
profitability has roughly the same power as book-to-
market predicting the cross section of returns. Profitable
firms generate higher average returns than unprofitable
firms. The second and third specifications replace gross
profitability with earnings-to-book equity and free cash
flow-to-book equity, respectively. These variables have
much less power than gross profitability. The fourth and
fifth specifications show that gross profitability subsumes
these other profitability variables. The sixth specification
shows that free cash flow subsumes earnings. The seventh
specification shows that free cash flow has incremental
power above that in gross profitability after controlling
for earnings, but that gross profitability is still the stron-
ger predictive variable.
Appendix A.2 performs similar regressions employing
alternative earnings variables. In particular, it considers
earnings before interest, taxes, depreciation, and amorti-
zation (EBITDA) and selling, general, and administrative
expenses (XSGA), which together represent a decomposi-
tion of gross profits. These regressions show that EBITDA-
to-assets and XSGA-to-assets each have significant power
predicting the cross section of returns, both individually
and jointly, but that gross profits-to-assets subsumes
their predictive powers. It also considers regressions
employing the DuPont model decomposition of gross
profits-to-assets into asset turnover (sales-to-assets, an
accounting measure of efficiency) and gross margins
(gross profits-to-sales, a measure of market power).
These variables also have power predicating the cross
section of returns, both individually and jointly, but
(footnote continued)
shareholder equity, I employ tiered definitions largely consistent with
those used by Fama and French (1993) to construct their high minus low
factor (HML). Stockholders equity is as given in Compustat (SEQ) if
available, or else common equity plus the carrying value of preferred
stock (CEQþ PSTX) if available, or else total assets minus total liabilities
(AT  LT). Deferred taxes is deferred taxes and investment tax credits
(TXDITC) if available, or else deferred taxes and/or investment tax credit
(TXDB and/or ITCB). Preferred stock is redemption value (PSTKR) if
available, or else liquidating value (PSTKRL) if available, or else carrying
value (PSTK).
4 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Table 1
Fama and MacBeth regressions of returns on measures of profitability.
Panel A reports results from Fama and MacBeth regressions of returns on gross profits (revenues minus cost of goods sold, REVT  COGS) scaled by
assets (AT), as well as income before extraordinary items (IB) and free cash flow (net income plus amortization and depreciation minus changes in
working capital and capital expenditures, NIþ DP  WCAPCH CAPX) each scaled by book equity. Panel B repeats the tests employing profitability
measures demeaned by industry (Fama and French 49). Regressions include controls for book-to-market [log(B/M)], size [log(ME)], and past performance
measured at horizons of one month ðr 1,0 Þ and 12 to two months ðr12,2 Þ. Independent variables are trimmed at the 1% and 99% levels. The sample excludes
financial firms (those with one-digit standard industrial classification codes of six) and covers July 1963 to December 2010.

Slope coefficients (102 ) and [test-statistics] from

regressions of the form r tj ¼ b0 xtj þ Etj

Independent (1) (2) (3) (4) (5) (6) (7)

variable

Panel A: Straight profitability variables

Gross profitability 0.75 0.69 0.62 0.61
[5.49] [5.22] [4.63] [4.59]
Earnings 0.22 0.08  0.02  0.07
[0.84] [0.31] [  0.06] [  0.27]
Free cash flow 0.27 0.20 0.39 0.33
[2.28] [1.64] [3.17] [2.67]
log(B/M) 0.35 0.30 0.26 0.34 0.30 0.27 0.31
[5.98] [4.97] [4.59] [5.54] [5.17] [4.48] [5.05]
log(ME)  0.09  0.12  0.13  0.11  0.11  0.13  0.11
[  2.29] [  3.24] [  3.20] [  2.78] [  2.80] [  3.34] [  2.92]
r 1,0  5.57  5.49  5.52  5.64  5.66  5.56  5.70
[  13.8] [  13.7] [  13.7] [  14.1] [  14.1] [  13.9] [  14.3]
r 12,2 0.76 0.78 0.78 0.74 0.74 0.76 0.73
[3.87] [4.02] [4.02] [3.80] [3.80] [3.93] [3.74]

Panel B: Profitability variables demeaned by industry

Gross profitability 1.00 0.94 0.94 0.92
[8.99] [9.75] [8.38] [8.39]
Earnings 0.28 0.04 0.09 0.02
[1.86] [0.15] [0.51] [0.13]
Free cash flow 0.16 0.09 0.25 0.21
[2.54] [1.23] [3.30] [2.54]

log(B/M) 0.35 0.32 0.30 0.33 0.34 0.31 0.34

[5.57] [5.40] [5.13] [5.15] [5.33] [5.12] [5.30]
log(ME)  0.09  0.11  0.11  0.10  0.10  0.11  0.10
[  2.19] [  2.65] [  2.64] [  2.62] [  2.31] [  2.69] [  2.35]
r 1,0  5.63  5.50  5.52  5.67  5.66  5.53  5.67
[  13.2] [  13.5] [  13.5] [  13.4] [  13.2] [  13.6] [  13.3]
r 12,2 0.76 0.78 0.78 0.75 0.75 0.77 0.74
[3.66] [3.98] [3.95] [3.63] [3.63] [3.92] [3.61]

gain lose their power when used in conjunction with
gross profitability. The analysis does suggest, however,
that high asset turnover primarily drives the high average
returns of profitable firms, while high gross margins are
the distinguishing characteristic of good growth stocks.
Gross profitability’s power predicting returns is not
driven by accruals or R&D expenditures. While these both
represent components of the wedge between earnings
and gross profits, and Sloan (1996) and Chan, Lakonishok,
and Sougiannis (2001) show, respectively, that these each
have power in the cross section, the results of Sloan and
Chan, Lakonishok, and Sougiannis cannot explain those
presented here. Appendix A.3 shows that gross profits-to-
assets retains power predicting returns after controlling
for accruals and R&D expenditures. This is not to say that
the results of Sloan and Chan, Lakonishok, and Sougiannis
do not exist independently, but simply that gross profit-
ability’s power to predict returns persists after controlling
for these earlier, well documented results.
Panel B repeats the tests of Panel A, employing gross
profits-to-assets, earnings-to-book equity and free cash
flow-to-book equity demeaned by industry. These tests
tell the same basic story, though the results here are even
stronger. Gross profits-to-assets is a powerful predictor of
the cross section of returns. The test-statistic on the slope
coefficient on gross profits-to-assets demeaned by indus-
try is more than one and a half times as large as that on
the variable associated with value (log(B/M)) and almost
two and a half times as large on the variable associated
with momentum ðr 12,2 Þ. Free cash flows also has some
power, though less than gross profits. Earnings convey
little information regarding future performance. The use
of industry-adjustment to better predict the cross section
of returns is investigated in greater detail in Section 4.
Because gross profitability appears to be the measure
of basic profitability with the most power predicting the
cross section of expected returns, it is the measure I focus
on for the remainder of the paper.
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28 5

Table 2
Excess returns to portfolios sorted on profitability.
This table shows monthly value-weighted average excess returns to portfolios sorted on gross profits-to-assets [(REVT  COGS)/AT], employing NYSE
breakpoints, and results of time series regressions of these portfolios’ returns on the Fama and French factors [the market factor (MKT), the size factor
small-minus-large (SMB), and the value factor high-minus-low (HML)], with test-statistics (in square brackets). It also shows time series average
portfolio characteristics [portfolio gross profits-to-assets (GP/A), book-to-market (B/M), average firm size (ME, in millions of dollars), and number of firms
(n)]. Panel B provides similar results for portfolios sorted on book-to-market. The sample excludes financial firms (those with one-digit standard
industrial classification codes of six) and covers July 1963 to December 2010.

Alphas and three-factor loadings Portfolio characteristics

e
Portfolio r a MKT SMB HML GP/A B/M ME n

Panel A: Portfolios sorted on gross profits-to-assets

Low 0.31  0.18 0.94 0.04 0.15 0.10 1.10 748 771
[1.65] [  2.54] [57.7] [1.57] [5.87]
2 0.41  0.11 1.03  0.07 0.20 0.20 0.98 1,100 598
[2.08] [  1.65] [67.5] [  3.13] [8.51]
3 0.52 0.02 1.02  0.00 0.12 0.30 1.00 1,114 670
[2.60] [0.27] [69.9] [  0.21] [5.42]
4 0.41 0.05 1.01 0.04  0.24 0.42 0.53 1,114 779
[1.94] [0.83] [70.6] [1.90] [  11.2]
High 0.62 0.34 0.92  0.04  0.29 0.68 0.33 1,096 938
[3.12] [5.01] [58.3] [  2.03] [  12.3]
High–low 0.31 0.52  0.03  0.08  0.44
[2.49] [4.49] [  0.99] [  2.15] [  10.8]

Panel B: Portfolios sorted on book-to-market

Low 0.39 0.13 0.98  0.09  0.39 0.43 0.25 1,914 965
[1.88] [2.90] [90.1] [  5.62] [  23.9]
2 0.45  0.02 0.99 0.05 0.04 0.31 0.54 1,145 696
[2.33] [  0.29] [78.1] [2.61] [2.23]
3 0.56 0.03 0.96 0.04 0.22 0.26 0.79 849 640
[2.99] [0.53] [63.5] [2.09] [9.71]
4 0.67  0.00 0.96 0.10 0.53 0.21 1.12 641 655
[3.58] [  0.03] [74.8] [5.66] [27.1]
High 0.80 0.07 1.01 0.25 0.51 0.21 5.47 367 703
[3.88] [1.04] [60.7] [10.7] [20.5]
High–low 0.41  0.06 0.03 0.34 0.91
[2.95] [  0.71] [1.44] [12.0] [30.0]

2.2. Sorts on profitability
The Fama and MacBeth regressions of Table 1 suggest
that profitability predicts average returns. These regres-
sions, because they weight each observation equally, put
tremendous weight on the nano- and micro-cap stocks,
which make up roughly two-thirds of the market by name
but less than 6% of the market by capitalization. The Fama
and MacBeth regressions are also sensitive to outliers and
impose a potentially misspecified parametric relation
between the variables, making the economic significance
of the results difficult to judge. This subsection attempts
to address these issues by considering the performance of
value-weighted portfolios sorted on profitability, non-
parametrically testing the hypothesis that profitability
predicts average returns.
Table 2 shows results of univariate sorts on gross
profits-to-assets [(REVT  COGS)/AT] and, for compari-
son, valuation ratios (book-to-market). The Spearman
rank correlation between gross profits-to-assets and
book-to-market ratios is  18%, and highly significant, so
strategies formed on the basis of gross profitability should
be growth strategies, while value strategies should hold
unprofitable firms. Portfolios are constructed using a
quintile sort, based on New York Stock Exchange (NYSE)
break points, and are rebalanced each year at the end of
June. The table shows the portfolios’ value-weighted
average excess returns, results of the regressions of the
portfolios’ returns on the three Fama and French factors,
and the time series average of the portfolios’ gross profits-
to-assets (GP/A), book-to-markets (B/M), and market
capitalizations (ME), as well as the average number of
firms in each portfolio (n). The sample excludes financial
firms (those with one-digit SIC codes of six) and covers
July 1963 to December 2010.3
3
Firms’ revenues, costs of goods sold, and assets are available on a
quarterly basis beginning in 1972 (Compustat data items REVTQ, COGSQ
and ATQ, respectively), allowing for the construction of gross profit-
ability strategies using more current public information than those
presented here. These high frequency gross profitability strategies are
even more profitable (see Appendix A.4). Despite these facts, I focus on
gross profitability measured using annual data. I am particularly inter-
ested in the persistent power gross profitability has predicting returns
and its relation to the similarly persistent value effect. While the high
frequency gross profitability strategy is most profitable in the months
immediately following portfolio formation, its profitability persists for
more than three years. Focusing on the strategy formed using annual
profitability data ensures that results are truly driven by the level of
profitability, and not surprises about profitability like those that drive
post earnings announcement drift. The low frequency profitability
strategy also incurs lower transaction costs, turning over only once
every four years, less frequently than the corresponding value strategy,
and only a quarter as often as the high frequency profitability strategy.
6 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28
The table shows that the gross profits-to-assets port-
folios’ average excess returns are generally increasing
with profitability, with the most profitable firms earning
0.31% per month higher average returns than the least
profitable firms, with a test-statistic of 2.49. This signifi-
cant profitable minus unprofitable return spread is
observed despite the fact that the strategy is a growth
strategy, with a large, significant negative loading on HML
(high minus low). As a result, the abnormal returns of the
profitable minus unprofitable return spread relative to the
Fama and French three-factor model is 0.52% per month,
with a test-statistic of 4.49.4
Consistent with the observed variation in HML load-
ings, the portfolios sorted on gross profitability exhibit
large variation in the value characteristic. Profitable firms
tend to be growth firms, in the sense of having low book-
to-markets, and unprofitable firms tend to be value firms,
with high book-to-market. In fact, the portfolios sorted on
gross profitability exhibit roughly half the variation in
book-to-markets as portfolios sorted directly on book-to-
market (Table 2, Panel B). Profitable firms also tend to be
growth firms, in the sense that they grow faster. Gross
profitability is a powerful predictor of future growth in
gross profitability, earnings, free cash flow, and payouts
(dividends plus repurchases) at both three- and 10-year
horizons (see Appendix A.5).
While the high gross profits-to-assets stocks resemble
typical growth firms in both characteristics and covar-
iances (low book-to-markets and negative HML loadings),
they are distinctly dissimilar in terms of expected returns.
That is, while they appear to be typical growth firms
under standard definitions, they are good growth firms,
exceptional in their tendency to outperform the market
despite their low book-to-markets.
Because the profitability strategy is a growth strategy,
it provides a great hedge for value strategies. The monthly
average returns to the profitability and value strategies
presented in Table 2 are 0.31% and 0.41% per month,
respectively, with standard deviations of 2.94% and 3.27%.
An investor running the two strategies together would
capture both strategies’ returns, 0.71% per month, but
would face no additional risk. The monthly standard
deviation of the joint strategy, despite having positions
twice as large as those of the individual strategies, is only
2.89%, because the two strategies’ returns have a correla-
tion of 0.57 over the sample. That is, while the 31 basis
point per month gross profitability spread is somewhat
modest, it is a payment an investor receives (instead of
pays) for insuring a value strategy. As a result, the test-
statistic on the average monthly returns to the mixed
profitability and value strategy is 5.87, and its realized
(footnote continued)
Using the annual data has the additional advantage of extending the
sample 10 years.
4
Including financial firms reduces the profitable minus unprofitable
return spread to 0.25% per month, with a test-statistic of 1.82, but
increases the Fama and French alpha of the spread to 0.61% per month,
with a test-statistic of 5.62. Most financial firms end up in the first
portfolio, because their large asset bases result in low profits-to-assets
ratios. This slightly increases the low profitability portfolio’s average
returns but also significantly increases its HML loading.
2 Profitability
Value
50/50 mix
Sharpe ratio (annualized)
1.5
1
0.5
0
−0.5
−1
1970 1975 1980 1985 1990 1995 2000 2005 2010
Fig. 1. Performance over time of profitability and value strategies.
The figure shows the trailing five-year Sharpe ratios of profitability
and value strategies (dashed and dotted lines, respectively) and a 50/50
mix of the two (solid line). The strategies are long–short extreme value-
weighted quintiles from sorts on gross profits-to-assets and book-to-
market, respectively, and correspond to the strategies considered in
Table 2. The sample excludes financial firms and covers June 1963 to
December 2010.
annual Sharpe ratio is 0.85, two and a half times the 0.34
observed on the market over the same period. The
strategy is orthogonal to momentum.
Fig. 1 shows the performance over time of the profit-
ability strategy presented in Table 2. The figure shows the
strategy’s realized annual Sharpe ratio over the preceding
five years at the end of each month between June 1968
and December 2010 (dashed line). It also shows the
performance of a similarly constructed value strategy
(dotted line) and a 50/50 mix of the two (solid line).
The figure shows that while both the profitability and
value strategies generally performed well over the sam-
ple, both had significant periods in which they lost
money. Profitability performed poorly from the mid-
1970s to the early-1980s and over the middle of the
2000s, while value performed poorly over the 1990s.
Profitability generally performed well in the periods when
value performed poorly, while value generally performed
well in the periods when profitability performed poorly.
As a result, the mixed profitability-value strategy never
had a losing five-year period over the sample (July 1963
to December 2010).
2.3. Profitability and size
The value-weighted portfolio results presented in
Table 2 suggest that the power that gross profits-to-
assets has predicting the cross section of average returns
is economically as well as statistically significant.
By analyzing portfolios double sorted on size and profit-
ability, this subsection shows that its power is economic-
ally significant even among the largest, most liquid stocks.
Portfolios are formed by independently quintile sorting on
the two variables (market capitalization and gross profits-
to-assets), using NYSE breaks. The sample excludes finan-
cial firms and covers July 1963 to December 2010.
Table 3 reports time series average characteristics of
the size portfolios. More than half of firms are in the
nano-cap portfolio, but these stocks make up less than 3%
of the market by capitalization, while the large cap
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28
Table 3
Size portfolio time series average characteristics.
This table reports the time series averages of the characteristics of quintile portfolios sorted on market equity. Portfolio break points are based on NYSE
stocks only. The sample excludes financial firms (those with one-digit standard industrial classification codes of six) and covers July 1963 to
December 2010.
Characteristic (Small)
Number of firms 2,427
Percent of firms 54.2
Average capitalization (millions of dollars) 39.6
Total capitalization (billions of dollars) 101
Total capitalization (percent) 2.43
Portfolio book-to-market 2.64
Portfolio gross profits-to-assets 0.27
portfolio typically contains fewer than 350 stocks but
makes up roughly three-quarters of the market by capi-
talization. The portfolios exhibit little variation in profit-
ability, but a great deal of variation in book-to-market,
with the smaller stocks tending toward value and the
larger stocks toward growth.
Table 4 reports the average returns to the portfolios
sorted on size and gross profits-to-assets. It also shows the
average returns of both sorts’ high minus low portfolios and
results of time series regressions of these high minus low
portfolios’ returns on the Fama and French factors. It also
shows the average number of firms in each portfolio, and the
average portfolio book-to-markets. Because the portfolios
exhibit little variation in gross profits-to-assets within profit-
ability quintiles, and little variation in size within size
quintiles, these characteristics are not reported.
The table shows that the profitability spread is large
and significant across size quintiles. While the spreads are
decreasing across size quintiles, the Fama and French
three-factor alpha is almost as large for the large-cap
profitability strategy as it is for small-cap strategies,
because the magnitudes of the negative HML loadings
on the profitability strategies are increasing across size
quintiles. That is, the predictive power of profitability is
economically significant even among the largest stocks,
and its incremental power above and beyond book-to-
market is largely undiminished with size.
Among the largest stocks, the profitability spread of 26
basis points per month (test-statistic of 1.88) is consider-
ably larger than the value spread of 14 basis points per
month (test-statistic of 0.95, untabulated). The large cap
profitability and value strategies have a negative correla-
tion of  0.58 and, consequently, perform well together.
While the two strategies’ realized annual Sharpe ratios
over the period are only 0.27 and 0.14, respectively, a
50/50 mix of the two strategies had a Sharpe ratio of 0.44.
While not nearly as large as the 0.85 Sharpe ratio on
the 50/50 mix of the value-weighted profitability and value
strategies that trade stocks of all size observed in Section 2,
this Sharpe ratio still greatly exceeds the 0.34 Sharpe ratio
observed on the market over the same period. It does so
despite trading exclusively in the Fortune 500 universe.
2.4. International evidence
The international evidence also supports the hypoth-
esis that gross profits-to-assets has roughly the same
7
(2) (3) (4) (Large)
749 484 384 335
16.9 11.3 9.26 8.25
206 509 1,272 9,494
173 273 544 3,652
3.73 6.13 12.6 75.1
1.36 1.06 0.88 0.61
0.28 0.26 0.25 0.27
power as book-to-market predicting the cross section of
expected returns. Table 5 shows results of univariate sorts
on gross profits-to-assets and book-to-market, like those
presented in Table 2, performed on stocks from developed
markets outside the US, including those from Australia,
Austria, Belgium, Canada, Denmark, Finland, France,
Germany, Great Britain, Hong Kong, Italy, Japan, the
Netherlands, New Zealand, Norway, Singapore, Spain,
Sweden, and Switzerland. The data come from Compustat
Global. The sample excludes financial firms and covers
July 1990 to October 2009. The table shows that the
profitability spread in international markets is significant
and even larger than the international value spread.
3. Profitability and value
The negative correlation between profitability and
book-to-market observed in Table 2 suggests that the
performance of value strategies can be improved by
controlling for profitability and that the performance of
profitability strategies can be improved by controlling for
book-to-market. A univariate sort on book-to-market
yields a value portfolio polluted with unprofitable stocks
and a growth portfolio polluted with profitable stocks.
A value strategy that avoids holding stocks that are more
unprofitable than cheap, and avoids selling stocks that are
more profitable than expensive, should outperform con-
ventional value strategies. Similarly, a profitability strat-
egy that avoids holding stocks that are profitable but fully
priced, and avoids selling stocks that are unprofitable but
nevertheless cheap, should outperform conventional prof-
itability strategies.
3.1. Double sorts on profitability and book-to-market
This subsection tests these predictions by analyzing
the performance of portfolios double sorted on gross
profits-to-assets and book-to-market. Portfolios are
formed by independently quintile sorting on the two
variables, using NYSE breaks. The sample excludes finan-
cial firms and covers July 1963 to December 2010. Table 6
shows the double sorted portfolios’ average returns, the
average returns of both sorts’ high minus low portfolios
and results of time series regressions of these high minus
low portfolios’ returns on the Fama and French factors.
It also shows the average number of firms in each
portfolio, and the average size of firms in each portfolio.
8 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Table 4
Double sorts on gross profits-to-assets and market equity.
This table shows the value-weighted average excess returns to portfolios double sorted, using NYSE breakpoints, on gross profits-to-assets and market
equity, and results of time series regressions of both sorts’ high minus low portfolios’ returns on the Fama and French factors [the market, size and value
factors MKT, SMB (small-minus-large), and HML (high-minus-low)]. Test statistics are given in square brackets. The table also shows the average number
of firms in each portfolio and each portfolios’ average book-to-market (the portfolios exhibit little gross-profits to asset variation within size quintiles and
little size variation within profitability quintiles). The sample excludes financial firms (those with one-digit standard industrial classification codes of six)
and covers July 1963 to December 2010.

Panel A: Portfolio average excess returns and time series regression results
Gross profits-to-asset quintiles Profitability strategies

e
Low 2 3 4 High r a bmkt bsmb bhml

Size quintiles
Small 0.40 0.64 0.78 0.89 1.07 0.67 0.63 0.05  0.13 0.13
[4.59] [4.27] [1.48] [  2.68] [2.47]
2 0.37 0.71 0.71 0.73 0.90 0.53 0.54 0.01 0.06  0.08
[3.97] [3.96] [0.35] [1.34] [  1.58]
3 0.40 0.73 0.74 0.68 0.81 0.41 0.38 0.10 0.18  0.15
[2.88] [2.71] [3.02] [4.01] [  3.05]
4 0.45 0.62 0.59 0.65 0.84 0.38 0.45 0.03 0.21  0.35
[2.82] [3.62] [1.00] [5.14] [  7.95]
Big 0.30 0.37 0.49 0.36 0.55 0.26 0.50  0.05  0.05  0.51
[1.88] [3.90] [  1.56] [  1.09] [  11.3]

Small-minus-big strategies
re 0.10 0.28 0.29 0.53 0.51
[0.39] [1.40] [1.45] [2.64] [2.37]
a  0.21  0.16  0.13  0.03  0.08
[  1.30] [  1.63] [  1.26] [  0.31] [  0.72]
bmkt  0.03  0.00  0.01 0.01 0.07
[  0.69] [  0.01] [  0.45] [0.53] [2.88]
bsmb 1.54 1.34 1.34 1.32 1.45
[28.9] [41.1] [38.9] [39.7] [41.9]
bhml  0.22 0.20 0.17 0.48 0.42
[  3.90] [5.56] [4.56] [13.5] [11.1]

Panel B: Portfolio average number of firms and portfolio book-to-markets

Gross profits-to-asset quintiles Gross profits-to-asset quintiles

Low 2 3 4 High Low 2 3 4 High

Number of firms Portfolio book-to-market

Size quintiles
Small 421 282 339 416 518 4.17 4.65 2.61 1.85 1.07
2 124 106 123 143 162 1.49 1.94 1.87 1.14 0.69
3 84 76 80 89 104 1.26 1.44 1.24 0.94 0.54
4 77 68 66 68 78 1.15 1.07 0.98 0.73 0.43
Big 63 64 60 62 72 0.97 0.82 0.92 0.42 0.27

Because the portfolios exhibit little variation in gross
profits-to-assets within profitability quintiles and little
variation in gross book-to-market within book-to-market
quintiles, these characteristics are not reported.
The table confirms the prediction that controlling for
profitability improves the performance of value strategies
and controlling for book-to-market improves the perfor-
mance of profitability strategies. The average value spread
across gross profits-to-assets quintiles is 0.68% per month
and in every book-to-market quintile exceeds the 0.41%
per month spread on the unconditional value strategy
presented in Table 2. The average profitability spread
across book-to-market quintiles is 0.54% per month and
in every book-to-market quintile exceeds the 0.31% per
month spread on the unconditional profitability strategy
presented in Table 2.
An interesting pattern also emerges in Panel B of the
table, which shows the number of firms in each portfolio
and the average size of firms in the portfolios. The table
shows that while more profitable growth firms tend to be
larger than less profitable growth firms, more profitable
value firms tend to be smaller than less profitable value
firms. So while little difference exists in size between
unprofitable value and growth firms, profitable growth
firms are large but highly profitable value firms are small.
Appendix A.6 presents results of similar tests per-
formed within the large and small cap universes, defined
here as stocks with market capitalization above and
below the NYSE median, respectively. These results are
largely consistent with the all-stock results presented in
Table 6.
3.2. Fortune 500 profitability and value strategies
Table 6 suggests that large return spreads can be
achieved by trading the corners of a double sort on value
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28 9

Table 5
Excess returns to portfolios sorted on profitability, international evidence.
This table shows monthly value-weighted average excess returns to portfolios of stocks from developed markets outside North America (Australia,
Austria, Belgium, Denmark, Finland, France, Germany, Great Britain, Hong Kong, Italy, Japan, the Netherlands, New Zealand, Norway, Singapore, Spain,
Sweden, and Switzerland) sorted on gross profits-to-assets [(REVT  COGS)/AT], employing NYSE breakpoints, and results of time series regressions of
these portfolios’ returns on the Fama and French factors [the market factor (MKT), the size factor small-minus-large (SMB), and the value factor high-
minus-low (HML)], with test-statistics (in square brackets). It also shows time series average portfolio characteristics [portfolio gross profits-to-assets
(GP/A), book-to-market (B/M), average firm size (ME, in millions of dollars), and number of firms (n)]. The sample excludes financial firms (those with
one-digit standard industrial classification codes of six) and covers July 1990 to October 2009.

Panel A: Portfolios sorted on gross profits-to-assets

Global three-factor loadings Portfolio characteristics

e
r a MKT SMB HML GP/A B/M ME n

Low  0.16  0.72 1.23 0.35 0.21 0.09 0.91 936 1,211
[  0.37] [  2.81] [20.8] [3.06] [1.98]
2 0.19  0.29 1.10 0.26 0.19 0.20 0.79 1,530 1,211
[0.50] [  1.65] [27.2] [3.26] [2.65]
3 0.29  0.07 1.12 0.20  0.12 0.32 0.72 1,843 1,211
[0.76] [  0.40] [29.7] [2.76] [  1.76]
4 0.44 0.17 0.94 0.06  0.03 0.54 0.95 1,946 1,211
[1.44] [1.30] [31.7] [1.05] [  0.65]
High 0.60 0.27 0.88 0.32 0.11 1.02 1.46 940 1,211
[1.95] [1.60] [22.7] [4.21] [1.58]

High–low 0.76 0.99  0.35  0.04  0.10

[2.25] [2.97] [  4.61] [  0.24] [  0.73]

Panel B: Portfolios sorted on book-to-market

Low 0.09  0.16 1.05 0.07  0.25 0.32 0.19 2,180 1,211
[0.25] [  1.09] [30.6] [1.04] [  4.11]
2 0.29  0.18 1.07 0.23 0.18 0.29 0.46 2,041 1,211
[0.81] [  1.16] [29.3] [3.24] [2.78]
3 0.29  0.21 1.06 0.17 0.29 0.27 0.71 1,465 1,211
[0.85] [  1.39] [30.9] [2.47] [4.73]
4 0.44  0.07 1.05 0.28 0.33 0.31 1.08 941 1,211
[1.25] [  0.37] [25.0] [3.49] [4.52]
High 0.61 0.15 1.01 0.38 0.28 0.40 8.13 564 1,211
[1.79] [0.84] [24.8] [4.83] [3.88]

High–low 0.51 0.31  0.03 0.31 0.53

[2.12] [1.46] [  0.69] [3.30] [6.11]

and profitability: Profitable value firms dramatically out-
perform unprofitable growth firms. While Section 2.3
already shows that the Sharpe ratio on the large cap
mixed value and growth strategy considered is 0.44, a
third higher than that on the market, this performance is
driven by the fact that the profitability strategy is an
excellent hedge for value. As a result, the large cap mixed
value and growth strategy has extremely low volatility
(standard deviations of monthly returns of 1.59%) and,
consequently, has a high Sharpe ratio despite generating
relatively modest average returns (0.20% per month). This
subsection shows that a simple trading strategy based on
gross profits-to-assets and book-to-market generates
average excess returns of almost 8% per year. It does so
despite trading only infrequently, in only the largest, most
liquid stocks.
The strategy I consider is constructed within the 500
largest nonfinancial stocks for which gross profits-to-
assets and book-to-market are both available. Each year
I rank these stocks on both their gross profits-to-assets
and book-to-market ratios, from 1 (lowest) to 500 (high-
est). At the end of each June the strategy buys one dollar
of each of the 150 stocks with the highest combined
profitability and value ranks, and it shorts one dollar of
each of the 150 stocks with the lowest combined ranks.5
The performance of this strategy is provided in Table 7.
The table also shows, for comparison, the performance of
similarly constructed strategies based on profitability and
value individually.
This simple strategy, which trades only liquid large cap
stocks, generates average excess returns of 0.62% per
month and has a realized annual Sharpe ratio of 0.74,
more than twice that observed on the market over the
same period. The strategy requires little rebalancing,
because both gross profits-to-assets and book-to-market
are highly persistent. Only one-third of each side of the
strategy turns over each year.
5
Well-known firms among those with the highest combined gross
profits-to-assets and book-to-market ranks at the end of the sample
(July through December of 2010) are Astrazeneca, GlaxoSmithKline, JC
Penney, Sears, and Nokia, and the lowest ranking firms include Ivanhoe
Mines, Ultra Petroleum, Vertex Pharmaceuticals, Marriott International,
Delta Airlines, Lockheed Martin, and Unilever. The largest firms held on
the long side of the strategy are WalMart, Johnson & Johnson, AT&T,
Intel, Verizon, Kraft, Home Depot, CVS, Eli Lilly, and Target, and the
largest firms from the short side are Apple, IBM, Philip Morris, McDo-
nald’s, Schlumberger, Disney, United Technologies, Qualcomm, Amazon,
and Boeing.
10 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Table 6
Double sorts on gross profits-to-assets and book-to-market.
This table shows the value-weighted average excess returns to portfolios double sorted, using NYSE breakpoints, on gross profits-to-assets and book-
to-market, as well results of time series regressions of both sorts’ high minus low portfolios’ returns on the Fama and French factors [the market, size and
value factors MKT, SMB (small-minus-large), and HML (high-minus-low)]. Test statistics are given in square brackets. The table also shows the average
number of firms, and the average size of firms, in each portfolio (the portfolios exhibit little gross-profits to asset variation within book-to-market
quintiles and little book-to-market variation within profitability quintiles). The sample excludes financial firms (those with one-digit standard industrial
classification codes of six) and covers July 1963 to December 2010.

Panel A: Portfolio average returns and time series regression results

Gross profits-to-asset quintiles Profitability strategies

e
Low 2 3 4 High r a bmkt bsmb bhml

Book-to-market quintiles
Low  0.08 0.19 0.27 0.26 0.56 0.64 0.83  0.24  0.27  0.01
[3.52] [4.76] [  6.03] [  4.81] [  0.18]
2 0.19 0.30 0.40 0.70 0.90 0.70 0.69  0.12 0.26 0.01
[4.13] [4.00] [  3.05] [4.61] [0.09]
3 0.38 0.39 0.74 0.69 0.87 0.49 0.27 0.09 0.53 0.10
[2.80] [1.64] [2.30] [9.89] [1.77]
4 0.50 0.60 0.94 1.04 0.93 0.43 0.28 0.07 0.65  0.14
[2.47] [2.06] [  0.94] [9.40] [  1.27]
High 0.65 0.83 0.96 1.09 1.08 0.44 0.34  0.04 0.51  0.08
[2.38] [1.79] [1.83] [12.6] [  2.52]

Value strategies
re 0.73 0.64 0.69 0.83 0.52
[3.52] [3.42] [3.76] [4.74] [2.81]
a 0.45 0.27 0.39 0.38  0.03
[2.76] [1.65] [2.26] [2.80] [  0.20]
bmkt  0.18  0.06  0.03  0.06 0.02
[  4.77] [  1.44] [  0.86] [  1.95] [0.72]
bsmb  0.04 0.27 0.32 0.74 0.75
[  0.75] [5.00] [5.57] [16.6] [16.2]
bhml 0.91 0.81 0.58 0.69 0.85
[15.7] [14.1] [9.52] [14.2] [17.0]

Panel B: Portfolio average number of firms and average firm size (millions of dollars)
Gross profits-to-asset quintiles Gross profits-to-asset quintiles

Low 2 3 4 High Low 2 3 4 High

Number of firms Average firm size

B/M quintiles
Low 195 101 128 194 343 653 1,461 1,891 2,694 2,493
2 104 95 130 170 192 1,002 1,729 1,590 1,238 669
3 113 104 128 145 142 1,003 1,402 1,266 534 277
4 144 129 128 128 118 955 1,118 630 268 187
High 174 151 135 120 108 568 424 443 213 102

While the joint profitability and value strategy gen-
erates almost half its profits on the long side (0.28% per
month more than the sample average for the high
portfolio, as opposed to 0.34% per month less for the
low portfolio), its real advantages over the straight value
strategy accrues only to investors that can short. The
unprofitable growth stocks, in addition to under-
performing the growth stocks as a whole by 19 basis
points per month, provide a better hedge for profitable
value stocks than growth stocks do for value. Value minus
growth strategies necessarily entail large HML loadings.
The long–short strategy based jointly on profitability and
value has less exposure to systematic risks and is, conse-
quently, less volatile.
Most of these benefits do not require shorting indivi-
dual unprofitable growth stocks, but can be captured by
shorting the market as a whole or selling market futures.
That is, the strategy of buying profitable large cap value
stocks has a highly significant information ratio relative to
the long side of the value strategy and the market
(abnormal returns of 18 basis points per month, with a
test-statistic of 3.46). The profitable value stocks, when
hedged using the market, have a Sharpe ratio of 0.75 and
earn excess returns of nearly 1% per month when run at
market volatility.
3.3. Conditional value and profitability factors
Table 6 also suggests that Fama and French’s HML
factor would be more profitable if it were constructed
controlling for profitability. This subsection confirms this
hypothesis explicitly. It also shows that a profitability
factor, constructed using a similar methodology, has a
larger information ratio relative to the three Fama and
French factors than does UMD (up minus down), the
momentum factor available from Ken French’s Data
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28 11

Table 7
Performance of large stock profitability and value strategies.
This table shows the performance of portfolios formed using only the 500 largest nonfinancial firms for which gross profits-to-assets (GP/A) and book-
to-market (B/M) are both available, and results of time series regressions of these portfolios’ returns on the Fama and French factors [the market factor
(MKT), the size factor small-minus-large (SMB), and the value factor high-minus-low (HML)], with test-statistics (in square brackets). Portfolios are
tertile sorted on GP/A (Panel A), B/M (Panel B), and the sum of the firms’ GP/A and B/M ranks within the sample (Panel C). The table also shows time
series average portfolio characteristics [portfolio GP/A, portfolio B/M, average firm size (ME, in billions of dollars), and number of firms (n)]. The sample
covers July 1963 to December 2010.

Alphas and three-factor loadings Portfolio characteristics

e
Portfolio r a MKT SMB HML GP/A B/M ME n

Panel A: Portfolios sorted on gross profits-to-assets

Low 0.38  0.15 1.02  0.04 0.22 0.12 1.02 5.93 150
[1.88] [  1.96] [55.1] [  1.38] [7.89]
2 0.57 0.00 1.13 0.11 0.08 0.31 0.86 7.82 200
[2.51] [0.06] [74.4] [5.08] [3.36]
High 0.65 0.25 1.02 0.08  0.18 0.64 0.41 9.20 150
[3.03] [3.84] [68.5] [3.69] [  7.81]
High–low 0.27 0.40 0.00 0.11  0.40
[2.33] [3.75] [0.17] [3.23] [  10.5]

Panel B: Portfolios sorted on book-to-market

Low 0.38 0.07 1.10 0.04  0.48 0.51 0.25 1.03 150
[1.56] [1.08] [69.5] [1.86] [  20.2]
2 0.52  0.00 1.07 0.07 0.08 0.34 0.58 7.27 200
[2.49] [  0.02] [79.7] [3.56] [3.91]
High 0.70 0.02 1.02 0.06 0.52 0.21 1.55 5.33 150
[3.60] [0.40] [73.3] [2.85] [24.9]
High–low 0.32  0.05  0.08 0.01 1.01
[2.16] [  0.64] [  4.41] [0.56] [37.0]

Panel C: Portfolios sorted on average gross profits-to-assets and book-to-market ranks

Low 0.19  0.20 1.12 0.01  0.27 0.22 0.45 7.72 150
[0.79] [  2.25] [53.2] [0.25] [  8.51]
2 0.59 0.10 1.01 0.02 0.09 0.38 0.68 8.83 200
[3.00] [1.95] [87.3] [1.32] [5.05]
High 0.81 0.17 1.08 0.15 0.29 0.45 1.21 5.87 150
[3.81] [2.80] [77.0] [7.62] [13.8]
High–low 0.62 0.37  0.04 0.14 0.56
[5.11] [3.67] [  1.73] [4.30] [15.8]

Library (http://mba.tuck.dartmouth.edu/pages/faculty/ken. firms in the same book-to-market decile (NYSE breaks).

french/data_library.html).
These conditional value and profitability factors are
constructed using the same basic procedure employed in
the construction of HML. HML is constructed as an equal
weighted mix of large and small cap value strategies.
Large and small cap are defined as firms with market
capitalizations above and below the NYSE median size,
respectively. Each of the value strategies is long (short)
stocks in the top (bottom) tertile of book-to-market by
NYSE breaks (i.e., have book-to-markets higher (lower)
than 70% of NYSE stocks). The returns to these book-to-
market-sorted portfolios are value-weighted and reba-
lanced at the end of June each year.
The conditional value and profitability factors are
constructed similarly, but instead of using a tertile sort
on book-to-market, they use either tertile sorts on book-
to-market within gross profitability deciles or tertile sorts
on gross profitability within book-to-market deciles. That
is, a firm is deemed a value (growth) stock if it has a book-
to-market higher (lower) than 70% of the NYSE firms in
the same gross profitability decile (NYSE breaks) and is
considered profitable (unprofitable) if it has a gross
profits-to-assets higher (lower) than 70% of the NYSE
Table 8 shows results of time series regressions employ-
ing these HML-like factors, HML9GP (HML conditioned on
gross profitability) and PMU9BM (profitable minus unpro-
fitable conditioned on book-to-market), over the sample
July 1963 to December 2010.
The first specification shows that controlling for profit-
ability does improve the performance of HML. HML9GP
generates excess average returns of 0.54% per month over
the sample, with a test-statistic of 5.01. This compares
favorably with the 0.40% per month, with a test-statistic
of 3.25, observed on HML. The second and third specifica-
tions show that HML9GP has an extremely large informa-
tion ratio relative to standard HML and the three Fama
and French factors (abnormal return test-statistics
exceeding 4.00). It is essentially orthogonal to momen-
tum, so also has a large information ratio relative to the
three Fama and French factors plus UMD.
The fourth specification shows that the profitability
factor constructed controlling for book-to-market is
equally profitable. PMU9BM generates excess average
returns of 0.48% per month, with a test-statistic of 5.35.
The fifth and sixth specifications show that PMU9BM has a
large information ratio relative to HML or the three Fama
12 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Table 8
High-minus-low factor (HML) constructed conditioning on gross profitability.
This table shows the performance of HML-like factors based on book-to-market within gross profitability deciles ðHML9GPÞ and gross profitability
within book-to-market deciles ðPMU9BMÞ. That is, a firm is deemed a value (growth) stock if it has a book-to-market higher (lower) than 70% of the NYSE
firms in the same gross profitability decile and is considered profitable (unprofitable) if it has a gross profits-to-assets higher (lower) than 70% of the
NYSE firms in the same book-to-market decile. The strategies exclude financial firms (those with one-digit standard industrial classification codes of six).
The table shows the factors’ average monthly excess returns, as well as time series regression of the strategies’ returns on the Fama and French factors
[the market factor (MKT), the size factor small-minus-large (SMB), and the value factor high-minus-low (HML)], with test-statistics (in square brackets).
The sample covers July 1963 to December 2010.

Dependent variable

HML9GP PMU9BM HML PMU

Independent (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
variable

Intercept 0.54 0.23 0.23 0.48 0.48 0.51 0.40  0.07 0.32 0.02
[5.01] [4.55] [4.52] [5.35] [5.23] [5.54] [3.25] [  1.31] [3.30] [0.57]
MKT  0.03  0.07
[  2.37] [  3.24]
SMB 0.05 0.03
[2.86] [1.12]
HML 0.77 0.77 0.02  0.01
[45.4] [43.0] [0.51] [  0.26]
HML9GP 1.04  0.33
[47.6] [  22.7]
PMU9BM  0.18 0.98
[  6.73] [57.2]
Adj. R2 78.4% 78.7% 0.0% 1.4% 79.9% 85.8%

and French factors. In fact, its information ratio relative to
the three Fama and French factors exceeds that of UMD
(abnormal return test-statistics of 5.54 and 5.11, respec-
tively). It is essentially orthogonal to momentum, so has a
similarly large information ratio relative to the three
Fama and French factors plus UMD.
The seventh and eighth specifications show that while
standard HML has a high realized Sharpe ratio over the
sample, it is inside the span of HML9GP and PMU9BM.
HML loads heavily on HML9GP (slope of 1.04) and garners
a moderate, though highly significant, negative loading on
PMU9BM (slope of  0.18). These loadings explain all of
the performance of HML, which has insignificant abnor-
mal returns relative to these two factors. Including the
market and SMB as explanatory variables has essentially
no impact on this result (untabulated).
The last two specifications consider an unconditional
profitability factor, constructed without controlling for
book-to-market. They show that this unconditional factor
generates significant average returns but is much less
profitable than the factor constructed controlling for
book-to-market. The unconditional factor is also inside
the span of HML9GP and PMU9BM. It is long real profit-
ability, with a 0.98 loading on PMU9BM, but short real
value, with a  0.33 loading on HML9GP, and these
loadings completely explain its average returns.
4. Profitability commonalities across anomalies
This section considers how well a set of alternative
factors, constructed on the basis of industry-adjusted
book-to-market, past performance, and gross profitability,
does pricing a wide array of anomalies. While I remain
agnostic here with respect to whether these factors are
associated with priced risks, they do appear to be useful in
identifying underlying commonalities in seemingly dis-
parate anomalies. The Fama and French model’s success
explaining long-run reversals can be interpreted in a
similar fashion. Even if one does not believe that the
Fama and French factors truly represent priced risk
factors, they certainly explain long-run reversals in the
sense that buying long-term losers and selling long-term
winners yields a portfolio long small and value firms and
short large and growth firms. An investor can largely
replicate the performance of strategies based on long-run
past performance using the right recipe of Fama and
French factors, and long-run reversals do not, conse-
quently, represent a truly distinct anomaly.
In much the same sense, regressions employing
industry-adjusted value, momentum and gross profitabil-
ity factors suggest that most earnings related anomalies
(e.g., strategies based on price-to-earnings, or asset turn-
over), and a large number of seemingly unrelated anoma-
lies (e.g., strategies based on default risk, or net stock
issuance), are just different expressions of these three
basic underlying anomalies, mixed in various proportions
and dressed up in different guises.
The three types of anomalies considered here are as
follows.
1. Anomalies related to the construction of the factors
themselves: Strategies sorted on size, book-to-market,
past performance, and gross profitability;
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28 13

2. Earnings-related anomalies: Strategies sorted on return Table 9

on assets, earnings-to-price, asset turnover, gross
margins, and standardized unexpected earnings; and
3. Anomalies considered by Chen, Novy-Marx, and Zhang
(2010): Strategies sorted on the failure probability
measure of Campbell, Hilscher, and Szilagyi (2008),
the default risk O-score of Ohlson (1980), net stock
issuance, asset growth, total accruals, and (not con-
sidered in Chen, Novy-Marx, and Zhang) the organiza-
tional capital based strategy of Eisfeldt and Papanikolaou
(2011).
Alternative factor average excess returns and Fama and French factor
loadings.
This table shows the returns to the factors based on book-to-market,
performance over the first 11 months of the preceding year, and gross
profitability scaled by book assets, in which each of these characteristics
are demeaned by industry (the Fama and French 49), and the resultant
factors are hedged for industry exposure [industry-adjusted high-
minus-low (HMLn), up-minus-down (UMDn), and profitable-minus-
unprofitable (PMUn)]. The table also shows results of regressions
of each factor’s abnormal returns on the three Fama and French factors
[the market factor (MKT), the size factor small-minus-large (SMB), and
the value factor high-minus-low (HML)] and the momentum factor
up-minus-down (UMD). Test-statistics are provided in square brackets.
The sample covers July 1973 to December 2010.

Time series regression results

4.1. Explanatory factors
e
Factor E½r  a MKT SMB HML UMD
The factors employed to price these anomalies are n
HML 0.47 0.25 0.02 0.10 0.42 0.01
formed on the basis of book-to-market, past performance, [6.42] [5.41] [2.07] [6.71] [26.6] [1.09]
and gross profitability. They are constructed using the UMDn 0.58 0.28  0.06  0.07  0.09 0.58
basic methodology employed in the construction of HML. [4.17] [6.51] [  6.74] [  5.27] [  6.58] [63.7]
Because Table 1 suggests that industry-adjusted gross PMUn 0.27 0.35  0.08  0.11  0.08 0.05
[4.88] [6.90] [  7.39] [  6.78] [  4.63] [4.56]
profitability has more power than straight gross profit-
ability predicting the cross section of expected returns,
and the literature has shown similar results for value and December 2010, with dates determined by the availability
momentum, the factor construction employs industry of the quarterly earnings data employed in the construc-
adjusted sorts, and the factors’ returns are hedged for tion of some of the anomaly strategies considered later.
industry exposure.6 Specifically, each of these factors is In fact, all three of the industry-adjusted factors have
constructed as an equal weighted mix of large and small Sharpe ratios exceeding those on any of the Fama and
cap strategies, in which large and small are defined by the French factors. Table 9 also shows that while the four
NYSE median market capitalization. The strategies are Fama and French factors explain roughly half of the
long (short) firms in the top (bottom) tertile by NYSE returns generated by HMLn and UMDn, they do not
breaks on the primary sorting variable. For the value significantly reduce the information ratios of any of the
factor, this is log book-to-market, demeaned by industry three factors. These factors are considered in greater
(the Fama and French 49). For the momentum and profit- detail in Appendix A.7.
ability factors, it is performance over the first 11 months
of the preceding year and gross profits-to-assets, both 4.2. Explaining anomalies
again demeaned by industry. Each stock position is
hedged for industry exposure, by taking an offsetting Table 10 shows the average returns to the 15 anomaly
position of equal magnitude in the corresponding stock’s strategies, as well as the strategies’ abnormal returns
value-weighted industry portfolio.7 The returns to each relative to both the standard Fama and French three-
portfolio are value-weighted and rebalanced either at the factor model plus UMD (hereafter referred to as the Fama
end of June (value and profitability strategies) or the end and French four-factor model), and the alternative four-
of each month (momentum strategy). factor model employing the market and industry-adjusted
The basic return properties of these factors–industry- value, momentum, and profitability factors (HMLn, UMDn,
adjusted high minus low (HMLn), up minus down (UMDn), and PMUn, respectively). Abnormal returns relative to the
and profitable minus unprofitable (PMUn)–are shown in model employing the four Fama and French factors plus
Table 9. All three factors generate highly significant the industry-adjusted profitability factor, and relative to
average excess returns over the sample, July 1973 to the three-factor model employing just the market and
industry-adjusted value and momentum factors, are pro-
6
Cohen and Polk (1998), Asness, Porter, and Stevens (2000), and
vided in Appendix A.8.
Novy-Marx (2009, 2011) all consider strategies formed on the basis of The first four strategies considered in the table inves-
industry-adjusted book-to-market. Asness, Porter, and Stevens (2000) tigate anomalies related directly to the construction of
also consider strategies formed on industry-adjusted past performance. the Fama and French factors and the profitability factor.
These papers find that strategies formed on the basis of industry-
The strategies are constructed by sorting on size (end-of-
adjusted book-to-market and past performance significantly outperform
their conventional counterparts. year market capitalization), book-to-market, performance
7
The assets employed in the construction of the strategies can be over the first 11 months of the preceding year, and
thought of simply as portfolios that hold an individual stock and take a industry-adjusted gross profitability-to-assets. All four
short position of equal magnitude in the stock’s industry. In practice the strategies are long–short extreme deciles of a sort on
long and short sides of the value, momentum, and profitability strategies
are fairly well balanced with respect to industry, because the corre-
the corresponding sorting variable, using NYSE breaks.
sponding sorting characteristics are already industry-adjusted, resulting Returns are value weighted, and portfolios are rebalanced
in 80% to 90% of the industry hedges netting on the two sides. at the end of July, except for the momentum strategy,
14 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Table 10
Anomaly strategy average excess returns and abnormal performance.
This table reports the average excess returns (E½r e ) to strategies formed by sorting on the variables used in factor construction (market capitalization,
book-to-market, performance over the first 11 months of the preceding year, and gross profits-to-assets demeaned by industry and hedged for industry
exposure); earnings related-variables [return on assets, earnings-to-price, asset turnover, gross margins, standardized unexpected earnings (SUE)]; and
failure probability, default risk (Ohlson’s O-score), net stock issuance, asset growth, total accruals, and organizational capital. Strategies are long–short
extreme deciles from a sort on the corresponding variable, employing NYSE breaks, and returns are value-weighted. Momentum, return on assets, return
on equity, SUE, failure, and default probability strategies are rebalanced monthly, while the other strategies are rebalanced annually, at the end of June.
Strategies based on variables scaled by assets exclude financial firms. The table also reports abnormal returns relative to the Fama and French four-factor
model ðaFF4 Þ and results of time series regressions of the strategies’ returns on the alternative four-factor model employing the market and industry-
adjusted value, momentum and profitability factors (HMLn, UMDn, and PMUn, respectively). Test-statistics are provided in square brackets. The sample
covers July 1973 through December 2010, and is determined by the availability of quarterly earnings data.

Alternative model abnormal returns and factor loadings

Sorting variable used E½r e  aFF4 a MKT HMLn UMDn PMUn

in strategy construction

Market equity 0.35  0.19 0.38  0.01 0.43 0.25  1.35

[1.53] [  1.60] [1.54] [  0.24] [2.83] [3.20] [  6.42]
Book-to-market 0.58 0.05  0.03  0.05 1.65  0.02  0.43
[3.13] [0.39] [  0.20] [  1.64] [17.5] [  0.41] [  3.33]
Prior performance 1.43 0.52  0.14 0.07 0.47 2.25 0.08
[4.28] [3.99] [  0.92] [2.06] [4.93] [45.6] [0.61]
Industry-adjusted 0.21 0.32  0.04  0.06  0.04 0.04 1.01
profitability [2.34] [4.01] [  0.61] [  3.95] [  0.86] [1.63] [16.8]
Return on assets 0.67 0.84  0.15  0.10 0.06 0.35 2.33
[2.81] [4.63] [  0.75] [  2.39] [0.49] [5.53] [13.6]
Return on equity 1.02 0.82 0.07  0.13 0.77 0.42 1.52
[4.47] [4.23] [0.32] [  2.82] [5.76] [6.09] [8.23]
Asset turnover 0.54 0.47  0.15 0.26 0.15  0.13 2.05
[2.92] [2.45] [  0.81] [6.95] [1.40] [  2.24] [13.5]
Gross margins 0.02 0.42 0.01  0.01  0.46  0.07 1.00
[0.15] [3.33] [0.05] [  0.39] [  4.97] [  1.38] [7.78]
SUE 0.69 0.54 0.36 0.06  0.31 0.62 0.30
[4.00] [3.66] [2.20] [1.88] [  3.08] [11.9] [2.13]
Failure probability 0.76 0.94  0.26  0.34  0.34 1.30 2.19
[2.09] [4.44] [  0.99] [  6.41] [  2.11] [15.8] [9.93]
Ohlson’s O-score 0.11 0.59 0.09  0.14  0.51 0.16 0.85
[0.58] [4.55] [0.48] [  3.62] [  4.50] [2.75] [5.40]
Net stock issuance 0.73 0.62 0.21  0.09 0.64 0.07 0.81
[5.15] [4.70] [1.49] [  3.06] [7.31] [1.49] [6.71]
Total accruals 0.37 0.37 0.39  0.13 0.28 0.05  0.38
[2.35] [2.32] [2.16] [  3.46] [2.53] [0.84] [  2.54]
Asset growth 0.70 0.30 0.23  0.11 1.06 0.13  0.18
[4.17] [2.12] [1.37] [  3.10] [10.4] [2.48] [  1.26]
Organizational capital 0.46 0.30 0.27  0.01 0.06 0.18 0.25
[3.73] [2.55] [1.94] [  0.46] [0.71] [4.06] [2.11]
Root mean squared
pricing error 0.67 0.54 0.22

which is rebalanced monthly. The profitability strategy is
hedged for industry exposure. The sample covers July
1973 through December 2010.
The second column of Table 10 shows the strategies’
average monthly excess returns. All the strategies, with
the exception of the size strategy, exhibit highly signifi-
cant average excess returns over the sample. The third
column shows the strategies’ abnormal returns relative to
the Fama and French four-factor model. The top two lines
show that the Fama and French four-factor model prices
the strategies based on size and book-to-market. It strug-
gles, however, with the extreme sort on past performance,
despite the fact that this is the same variable used in the
construction of UMD. This reflects, at least partly, the fact
that selection into the extreme deciles of past performance
are little influenced by industry performance. The stan-
dard momentum factor UMD, which is constructed using
the less aggressive tertile sort, is formed more on the basis
of past industry performance. It consequently exhibits
more industry-driven variation in returns and looks less
like the decile sorted momentum strategy. The Fama and
French model hurts the pricing of the profitability based
strategy, because it is a growth strategy that garners a
significant negative HML loading despite generating sig-
nificant positive average returns. The fourth column
shows that the alternative four-factor model prices all
four strategies. Unsurprisingly, it prices the momentum
and profitability strategies with large significant loadings
on the corresponding industry-adjusted factors. It prices
the value strategy with a large significant loading on the
industry-adjusted value factor and a significant negative
loading on the industry-adjusted profitability factor. The
conventional value strategy is, as in Table 8, long real
value (HMLn) but short profitability.
The next five lines of Table 10 investigate earnings-
related anomalies. These strategies are constructed by
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28
sorting on return on assets, earnings-to-price, asset turn-
over, gross margins, and standardized unexpected earn-
ings. They are again long–short extreme deciles of a sort
on the corresponding sorting variable, using NYSE breaks.
The return-on-assets, asset turnover, and gross margin
strategies exclude financial firms (i.e., those with one-
digit SIC codes of six). Returns are value weighted. The
asset turnover and gross margin strategies are rebalanced
at the end of June; the others are rebalanced monthly.
The second column shows the strategies’ average
monthly excess returns. All of the strategies, with the
exception of that based on gross margins, exhibit highly
significant average excess returns over the sample. The
third column shows that the standard Fama and French
four-factor model performs extremely poorly pricing
earnings-related anomalies. This is admittedly tautologi-
cal, as the Fama and French model’s failure to price a
strategy is used here as the defining characteristic of an
anomaly.
The fourth column shows that the alternative four-
factor model explains the returns to all of the strategies,
with the exception of post earnings announcement drift,
for which the model can explain about half the excess
returns. All of the strategies have large, significant load-
ings on PMUn, especially the return on assets, earnings-to-
price, and asset turnover strategies. The fact that the
model prices the return on assets strategy is especially
remarkable, given that the strategy produces significant
returns only when rebalanced monthly using the most
recently available earnings information, while the profit-
ability factor is rebalanced annually employing only
relatively stale gross profitability information. The model
also does well pricing the strategy based on gross mar-
gins, even though the high margin firms tend to be
growth firms, a fact that drives the strategy’s large Fama
and French alpha, because the high margin firms also tend
to be profitable. The resulting large positive PMUn loading
effectively offsets the pricing effect of the large negative
HMLn loading.
The last six strategies in Table 10 are those considered,
along with value, momentum, and post earnings
announcement drift, by Chen, Novy-Marx, and Zhang
(2010). These strategies are based on the failure prob-
ability measure of Campbell, Hilscher, and Szilagyi (2008),
the default risk O-score of Ohlson (1980), net stock
issuance, asset growth, total accruals, and (not considered
in Chen, Novy-Marx, and Zhang) the Eisfeldt and
Papanikolaou (2011) organizational capital-based strat-
egy.8 All six anomalies are constructed as long–short
extreme decile strategies, and portfolio returns are
value-weighted. The strategies based on failure
8
Eisfeldt and Papanikolaou (2011) construct this strategy on their
accounting-based measure, which accumulates selling, general, and
administrative expenses (XSGA), the accounting variable most likely to
include spending on the development of organizational capital. The
stock of organizational capital is assumed to depreciate at a rate of 15%
per year, and the initial stock is assumed to be 10 times the level of
selling, general, and administrative expenses that first appear in the
data, though results employing this measure are not sensitive to these
choices. The trading strategy is formed by sorting on the organizational
capital measure within industries.
15
probability and Ohlson’s O-score are rebalanced monthly,
while the other four strategies are rebalanced annually, at
the end of June.
The second and third columns of Table 10 show the six
strategies’ average monthly excess returns and their
abnormal returns relative to the Fama and French four-
factor model. All of the strategies exhibit highly signifi-
cant average excess returns and four-factor alphas over
the sample. The fourth column shows that the four-factor
model employing the market and industry-adjusted HML,
UMD, and PMU explains the performance of all the
strategies except for that based on total accruals. The
model explains the poor performance of the failure prob-
ability and default probability firms primarily through
large, significant loadings on the industry-adjusted profit-
ability factor. Firms with low industry-adjusted gross
profits-to-assets tend to be firms that both the
Campbell, Hilscher, and Szilagyi (2008) and Ohlson
(1980) measures predict are more likely to default, and
this drives the performance of both strategies. The fact
that the model performs well pricing these two strategies
is again especially remarkable given that these anomalies
exist only at the monthly frequency, in the sense that
strategies based on the same sorting variables do not
produce significant excess returns when rebalanced
annually. The model explains the net stock issuance
anomaly primarily through loadings on HMLn and PMUn.
The low returns to net issuers are explained by the fact
that issuers tend to be industry-adjusted growth stocks
with low industry-adjusted profitability. The model
explains the out-performance of high organizational capi-
tal firms primarily through a positive loading PMUn,
suggesting that firms with large stocks of organizational
capital, at least as quantified by the Eisfeldt and
Papanikolaou (2011) measure, are more profitable than
those with small stocks of organizational capital. Direct
investigation of portfolios underlying organizational capi-
tal strategy confirms this prediction. Decile portfolios
sorted on organizational capital show strong monotonic
variation in gross profitability.
The alternative four-factor model also performs well in
the sense that it dramatically reduces the strategies’ root
mean squared pricing error. The root mean squared
average excess return across the 15 anomalies is 0.67%
per month. The root mean squared pricing error relative
to the alternative four-factor model is only 0.22% per
month, less than half the 0.54% per month root mean
squared pricing errors observed relative to the standard
Fama-French four-factor model. Appendix A.8 shows that
roughly two-thirds of the alternative four-factor model’s
improved performance relative to the standard Fama and
French four-factor model is due to the inclusion of the
industry-adjusted profitability factor, and one-third is due
to the industry-adjustments to the value and momentum
factors.
5. Conclusion
Gross profitability represents the other side of value.
The same basic philosophy underpins strategies based on
both valuation ratios and profitability. Both are designed
16 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

to acquire productive capacity cheaply. Value strategies
do this by financing the purchase of inexpensive assets
through the sale of expensive assets, and profitability
strategies achieve the same end by financing the purchase
of productive assets through the sale of unproductive
assets. Both strategies generate significant abnormal
returns.
While profitability is another dimension of value, and
gross profits-to-assets has roughly the same power as
book-to-market predicting the cross section of average
returns, profitable firms are extremely dissimilar from
value firms. Profitable firms generate significantly higher
average returns than unprofitable firms despite having, on
average, lower book-to-markets and higher market capi-
talizations. That is, while trading on gross profits-to-
assets exploits a value philosophy, the resulting strategy
is a growth strategy as measured by either characteristics
(valuation ratios) or covariances (HML loadings). Because
the value and profitability strategies’ returns are nega-
tively correlated, the two strategies work extremely well
together. In fact, a value investor can capture the full
profitability premium without taking on any additional
risk. Adding a profitability strategy on top of an existing
value strategy reduces overall portfolio volatility, despite
doubling the investor’s exposure to risky assets. Value
investors should consequently pay close attention to gross
profitability when selecting their portfolio holdings,
because controlling for profitability dramatically increases
the performance of value strategies.
These facts are difficult to reconcile with the inter-
pretation of the value premium provided by Fama and
French (1993), which explicitly relates value stocks’ high
average returns to their low profitabilities. In particular,
they note that ‘‘low-BE/ME firms have persistently high
earnings and high-BE/ME firms have persistently low
earnings,’’ suggesting that ‘‘the difference between the
returns on high- and low-BE/ME stocks, captures variation
through time in a risk factor that is related to relative
earnings performance’’ (p. 53). While a sort on book-to-
market does yield a value strategy that is short profit-
ability, a direct analysis of profitability shows that the
value premium is emphatically not driven by unprofitable
stocks.
My results present similar problems for the operating
leverage hypothesis of Carlson, Fisher, and Giammarino
(2004), which formalizes the intuition in Fama and French
(1993) and drives the value premium in Zhang (2005) and
Novy-Marx (2009, 2011). Under this hypothesis operating
leverage magnifies firms’ exposures to economic risks,
because firms’ profits look like levered claims on their
revenues. In models employing this mechanism, however,
operating leverage, risk, and expected returns are gener-
ally all decreasing with profitability, suggesting profitable
firms should under-perform unprofitable firms. This is
contrary to the relation between profitability and average
returns observed in the data.
The fact that profitable firms earn significantly higher
average returns than unprofitable firms also poses diffi-
culties for the Lettau and Wachter (2007) duration-based
explanation of the value premium. In their model, short-
duration assets are riskier than long-duration assets.
Value firms have short durations and, consequently,
generate higher average returns than longer duration
growth firms. In the data, however, gross profitability is
associated with long run growth in profits, earnings, free
cash flows, and dividends. Profitable firms, consequently,
have longer durations than less profitable firms, and the
Lettau and Wachter model, therefore, predicts, counter-
factually, that profitable firms should under-perform
unprofitable firms.
Appendix A
A.1. Correlations between variables employed in the Fama
and MacBeth regressions
Table A1 reports the time series averages of the
cross sectional Spearman rank correlations between the
independent variables employed in the Fama and Mac-
Beth regressions of Table 1. The table shows that the
earnings-related variables are, not surprisingly, all posi-
tively correlated with each other. Gross profitability and

Table A1
Spearman rank correlations between independent variables.
This table reports the time series averages of the cross section Spearman rank correlations between the independent variables employed in the Fama
and MacBeth regressions of Table 1: gross profitability [(REVT  COGS)/A], earnings (IB/A), free cash flow ((NI þDP  WCAPCH CAPX)/A), book-to-
market, market equity, and past performance measured at horizons of one month (r 1,0 ) and 12 to two months (r 12,2 ). Test-statistics are provided in
square brackets. The sample excludes financial firms (those with one-digit standard industrial classification codes of six) and covers 1963–2010.

Variable IB/A FCF/A B/M ME r 12,2 r1,0

Gross profitability (GP/A) 0.45 0.31  0.18  0.03 0.09 0.02

[58.7] [17.9] [  17.2] [  2.51] [7.00] [1.87]
Earnings (IB/A) 0.60  0.25 0.36 0.23 0.07
[16.7] [  8.89] [30.4] [14.4] [6.27]
Free cash flow (FCF/A)  0.03 0.20 0.17 0.07
[  1.33] [10.7] [10.6] [7.12]
Book-to-market (B/M)  0.26  0.09 0.02
[  13.1] [  4.98] [1.37]
Market equity (ME) 0.26 0.13
[11.3] [9.20]
Prior year’s performance (r 12,2 ) 0.08
[5.15]
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28 17

Table A2
Fama and MacBeth regressions employing EBITDA and XSGA.
This table reports results from Fama and MacBeth regressions of firms’ returns on gross profits (revenues minus cost of goods sold, Compustat REVT  COGS),
earnings before interest, taxes, depreciation, and amortization (EBITDA), and selling, general, and administrative expenses (XSGA), each scaled by assets (A).
Regressions include controls for book-to-market [log(B/M)], size [log(ME)], and past performance measured at horizons of one month (r 1,0 ) and 12 to two
months (r 12,2 ). Independent variables are trimmed at the 1% and 99% levels. The sample excludes financial firms (those with one-digit standard industrial
classification codes of six) and covers 1963–2010.

Slope coefficients (102 ) and [test-statistics]

from regressions of the form r tj ¼ b0 xtj þ Etj

Independent (1) (2) (3) (4) (5) (6)

variable

Gross profitability 0.75 0.53 1.47

[5.49] [3.76] [4.51]
EBITDA-to-assets 1.42 1.07 1.65
[4.28] [3.03] [4.73]
XSGA-to-assets 0.65  0.69 0.73
[4.07] [  2.05] [4.70]
log(B/M) 0.35 0.33 0.38 0.37 0.38 0.38
[5.98] [5.70] [6.94] [6.38] [6.49] [6.46]
log(ME)  0.09  0.13  0.07  0.11  0.10  0.10
[  2.29] [  3.44] [  1.69] [  2.99] [  2.65] [  2.78]
r 12,2 0.76 0.76 0.77 0.72 0.71 0.72
[3.87] [3.89] [3.96] [3.74] [3.71] [3.78]
r 1,0  5.57  5.57  5.67  5.71  5.83  5.85
[  13.8] [  14.1] [  14.0] [  14.4] [  14.6] [  14.8]

earnings are also negatively correlated with book-to-
market, with magnitudes similar to the negative correla-
tion observed between book-to-market and size. Earnings
and free cash flows are positively associated with size
(more profitable firms have higher market values), but
surprisingly the correlation between gross profitability
and size is negative, though weak. These facts suggest that
strategies formed on the basis of gross profits-to-assets
will be growth strategies and relatively neutral with
respect to size.
A.2. Tests employing other earnings variables
Earnings before interest, taxes, depreciation, and
amortization is gross profits minus operating expenses,
which largely consist of selling, general, and administra-
tive expenses. Table A2 shows results of Fama and
MacBeth regressions employing gross profits-to-assets
and a decomposition of gross profits-to-assets into
EBITDA-to-assets and XSGA-to-assets. EBITDA-to-assets
and XSGA-to-assets have time series average cross sec-
tional Spearman rank correlations with gross profits-to-
assets of 0.51 and 0.77, respectively, and are essentially
uncorrelated with each other. The table shows that both
variables have power explaining the cross section of
average returns, either individually or jointly. The table
also shows that while XSGA-to-assets has no power to
predict returns in regressions that include gross profits-
to-assets, EBITDA-to-assets retains incremental power
after controlling for gross profitability. Because of the
collinearity resulting from the fact that EBITDA-to-assets
and XSGA-to-assets together make up gross profits-to-
assets, all three variables cannot be used in the same
regression.
Gross profitability is also driven by two dimensions,
asset turnover and gross margins:
Gross profits Sales Gross profits
¼  , ð2Þ
Assets Assets
|fflfflffl{zfflfflffl} Sales
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
Asset turnover Gross margins
a decomposition known in the accounting literature as
the Du Pont model. Asset turnover, which quantifies the
ability of assets to generate sales, is often regarded as a
measure of efficiency. Gross margins, which quantifies
how much of each dollar of sales goes to the firm, is a
measure of profitability. It relates directly, in standard
oligopoly models, to firms’ market power. Asset turnover
and gross margins are generally negatively related. A firm
can increase sales and, thus, asset turnover, by lowering
prices, but lower prices reduce gross margins. Conversely,
a firm can increase gross margins by increasing prices, but
this generally reduces sales and, thus, asset turnover.9
Given this simple decomposition of gross profitability
into asset turnover and gross margins, it seems natural to
ask which of these two dimensions of profitability, if
either, drives profitability’s power to predict the cross
section of returns. The results suggest that both dimen-
sions have power, but that this power is subsumed by
basic profitability. That is, it appears that the decomposi-
tion of profitability into asset turnover and gross margins
does not add any incremental information beyond that
9
The time series average of the Spearman rank correlation of firms’
asset turnovers and gross margins in the cross section is  0.27, in the
sample spanning 1963–2010 that excludes financial firms. Both asset
turnover and gross margins are strongly positively correlated with gross
profitability in the cross section (time series average Spearman rank
correlations of 0.67 and 0.43, respectively).
18 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Table A3
Fama and MacBeth regressions with asset turnover and gross margins.
This table reports results from Fama and MacBeth regressions of firms’ returns on profitability [gross profits-to-assets, measured as revenues minus
cost of goods sold (REVT  COGS) scaled by assets (A)], asset turnover (REVT/A), and gross margins (GP/REVT). Regressions include controls for book-to-
market [log(B/M)], size [log(ME)], and past performance measured at horizons of one month (r 1,0 ) and 12 to two months (r12,2 ). Independent variables are
trimmed at the 1% and 99% levels. The sample excludes financial firms (those with one-digit standard industrial classification codes of six) and covers
1963–2010.

Slope coefficients (102 ) and [test-statistics]

from regressions of the form rtj ¼ b0 xtj þ Etj

Independent (1) (2) (3) (4) (5) (6) (7)

variable

Profits-to-assets 0.75 0.83 0.81 0.82

[5.49] [6.92] [5.70] [5.08]
Asset turnover 0.13  0.04 0.16  0.01
[2.59] [  0.87] [3.27] [  0.28]
Gross margins 0.28 0.12 0.37 0.15
[2.67] [0.97] [3.31] [0.83]
log(B/M) 0.35 0.31 0.33 0.36 0.37 0.33 0.37
[5.98] [5.32] [5.73] [6.09] [6.30] [5.70] [6.35]
log(ME)  0.09  0.10  0.11  0.10  0.09  0.10  0.09
[  2.29] [  2.44] [  2.66] [  2.35] [  2.29] [  2.39] [  2.26]
r 12,2  5.57  5.56  5.44  5.62  5.59  5.59  5.65
[  13.8] [  13.8] [  13.4] [  14.0] [  13.8] [  13.9] [  14.0]
r 1,0 0.76 0.77 0.83 0.75 0.76 0.78 0.75
[3.87] [3.94] [4.21] [3.85] [3.85] [3.98] [3.84]

contained in gross profitability alone. The results do
suggest, however, that high asset turnover is more
directly associated with higher returns, and high margins
are more strongly associated with good growth. That is,
high sales-to-assets firms tend to outperform on an
absolute basis, and firms that sell their goods at high
markups tend to be growth firms that outperform
their peers.
Table A3 shows results of Fama and MacBeth regres-
sions of firms’ returns on gross profitability, asset turn-
over, and gross margins. These regressions include
controls for book-to-market [log(B/M)], size [log(ME)],
and past performance measured at horizons of one month
ðr 1,0 Þ and 12 to two months ðr 12,2 Þ. Independent variables
are trimmed at the 1% and 99% levels. The sample covers
July 1963 to December 2010, and it excludes financial
firms (those with one-digit SIC codes of six).
Specification 1, which employs gross profitability, is
identical to the first specification in Table 1. It shows the
baseline result, that gross profitability has roughly the
same power as book-to-market predicting the cross sec-
tion of returns. The second and third specifications
replace gross profitability with asset turnover and gross
margins, respectively. Each of these variables has power
individually, but less power than gross profitability. The
fourth and fifth specifications show that gross margins
subsume either asset turnover or gross margins, but that
including asset turnover increases the coefficient esti-
mated on gross profitability and improves the precision
with which it is estimated. The sixth and seventh speci-
fications show that asset turnover and gross margins both
have power when used together, but neither has power
when used in conjunction with gross profitability.
Table A4 shows results of univariate sorts on asset
turnover and gross margins. These tests employ the same
methodology as that employed in Table 2, replacing gross
profitability with asset turnover and gross margins. The
table shows the portfolios’ value-weighted average excess
returns, results of time series regressions of the portfolios’
returns on the three Fama and French factors, and the
time series averages of the portfolios’ gross profits-to-
assets (GP/A), book-to-markets (B/M), and market capita-
lizations (ME), as well as the average number of firms in
each portfolio (n).
Panel A provides results for the five portfolios sorted
on asset turnover. The portfolios’ average excess returns
are increasing with asset turnover but show little varia-
tion in loadings on the three Fama and French factors. As a
result, the high minus low turnover strategy produces
significant average excess returns that cannot be
explained by the Fama and French model. The portfolios
show a great deal of variation in gross profitability, with
more profitable firms in the high asset turnover portfolios.
They show some variation in book-to-market, with the
high turnover firms commanding higher average valua-
tion ratios, but this variation in book-to-market across
portfolios is not reflected in the portfolios’ HML loadings.
Panel B provides results for the five portfolios sorted
on gross margins. Here the portfolios’ average excess
returns exhibit little variation across portfolios, but large
variation in their loadings on SMB and especially HML,
with the high margin firms covarying more with large
growth firms. As a result, while the high minus low
turnover strategy does not produce significant average
excess returns, it produces highly significant abnormal
returns relative to the Fama and French model, 0.37% per
month with a test-statistic of 4.35. The portfolios show
less variation in gross profitability than do the portfolios
sorted on asset turnover, though the high margin firms
are more profitable, on average, than the low margin
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28 19

Table A4
Excess returns to portfolios sorted on asset turnover and gross margins.
This table shows monthly value-weighted average excess returns to portfolios sorted, using NYSE breaks, on asset turnover (REVT/AT, Panel A) and
gross margins [(REVT  COGS)/REVT, Panel B]. It also shows results of time series regressions of these portfolios’ returns on the Fama and French factors
and time series average portfolio characteristics [portfolio gross profits-to-assets (GP/A), book-to-market (B/M), average firm size (ME, in millions of
dollars), and number of firms (n)]. Test-statistics are provided in square brackets. The sample excludes financial firms (those with one-digit standard
industrial classification codes of six) and covers 1963–2010.

Alphas and three-factor loadings Portfolio characteristics

re a MKT SMB HML GP/A B/M ME n

Panel A: Portfolios sorted on asset turnover

Low 0.29  0.11 0.91  0.05 0.00 0.16 0.93 1,004 821
[1.52] [  1.40] [50.1] [  2.10] [0.16]
2 0.45 0.09 1.00  0.07  0.15 0.25 0.71 1,461 671
[2.25] [1.51] [73.1] [  3.55] [  7.33]
3 0.51 0.10 1.01 0.00  0.08 0.34 0.79 1,336 664
[2.54] [1.94] [87.7] [0.01] [  4.9]
4 0.56 0.09 1.01 0.04 0.01 0.37 0.61 889 742
[2.77] [1.64] [75.8] [2.36] [0.32]
High 0.62 0.15 0.96 0.17  0.00 0.48 0.57 623 856
[3.01] [1.81] [49.2] [6.36] [  0.08]
High–low 0.34 0.26 0.04 0.23  0.01
[2.51] [1.94] [1.41] [5.17] [  0.14]

Panel B: Portfolios sorted on gross margins

Low 0.44  0.17 1.03 0.28 0.18 0.16 1.08 525 812
[2.06] [  2.89] [74.9] [14.9] [8.48]
2 0.50  0.04 0.99 0.04 0.20 0.26 0.87 1,119 623
[2.54] [  0.56] [63.9] [1.84] [8.50]
3 0.48  0.03 1.04  0.02 0.13 0.27 1.04 990 628
[2.43] [  0.49] [85.5] [  1.37] [7.09]
4 0.47 0.05 0.97  0.02  0.00 0.29 0.68 1,017 712
[2.49] [0.95] [80.3] [  1.45] [  0.21]
High 0.45 0.20 0.94  0.10  0.35 0.36 0.45 1,528 945
[2.29] [4.39] [88.6] [  6.75] [  21.8]
High–low 0.01 0.37  0.09  0.39  0.52
[0.13] [4.35] [  4.48] [  13.9] [  17.6]

firms. The portfolios sorted on gross margins exhibit far
more variation in book-to-market, however, than the
asset turnover portfolios, with high margin firms com-
manding high valuation ratios. These firms are emphati-
cally growth firms, both possessing the defining
characteristic (low book-to-markets) and garnering large
negative loadings on the standard value factor. These
growth firms selected on the basis of gross margins are
good growth firms, however, which dramatically outper-
form their peers in size and book-to-market.
A.3. Controlling for accruals and R&D
Accruals and R&D expenditures both represent com-
ponents of the wedge between earnings and gross profits.
Sloan (1996) shows that accruals have power predicting
the cross section or returns, hypothesizing that ‘‘if inves-
tors naively fixate on earnings, then they will tend to
overprice (underprice) stocks in which the accrual com-
ponent is relatively high (low)y[so] a trading strategy
taking a long position in the stock of firms reporting
relatively low levels of accruals and a short position in the
stock of firms reporting relatively high levels of accruals
generates positive abnormal stock returns’’ (p. 292). Chan,
Lakonishok, and Sougiannis (2001) provide a similar
result for R&D expenditures, showing that ‘‘companies
with high R&D to equity market value (which tend to
have poor past returns) earn large excess returns’’
(p. 2431).
Table A5 confirms these results independently but
shows that they are basically unrelated to the power
gross profitability has predicting returns. The table pre-
sents results from a series of Fama and MacBeth regres-
sions, similar to those presented in Table 1, employing
gross profits-to-assets, accruals, and R&D expenditures-
to-market, and controls for valuation ratios, size, and past
performance. Accruals are defined, as in Sloan (1996), as
the change in non-cash current assets, minus the change in
current liabilities (excluding changes in debt in current
liabilities and income taxes payable), minus depreciation.10
The table shows that while accruals and R&D expenditures
both have power predicting returns, these variables do not
explain the power gross profitability has predicting returns.
10
Specifically, this is defined as the change in Compustat annual
data item ACT (current assets), minus CHECH (change in cash/cash
equivalents), minus the change in LCT (current liabilities), plus the
change in DLC (debt included in liabilities), plus the change in TXP
(income taxes payable), minus DP (depreciation and amortization).
Variables are assumed to be publicly available by the end of June in
the calendar year following the fiscal year with which they are
associated. Following Sloan, accruals are scaled by average assets,
defined as the mean of current and prior year’s total assets (Compustat
data item AT).
20 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Table A5
Fama and MacBeth regressions with accruals and research and development (R&D) expenditures.
This table reports results from Fama and MacBeth regressions of firms’ returns on gross profits (REVT  COGS) scaled by assets (AT), accruals (DACT 
CHECH  DLCTþ DDCT þ DTXPDP) scaled by average assets (AT/2 þ lag(AT)/2), and research and development expenditures (XRD) scaled by market
value. Regressions include controls for book-to-market [log(B/M)], size [log(ME)], and past performance measured at horizons of one month (r 1,0 ) and 12
to two months (r 12,2 ). Independent variables are trimmed at the 1% and 99% levels. The sample excludes financial firms (those with one-digit standard
industrial classification codes of six) and covers July 1973 to December 2010, a period determined by the availability of high quality data on R&D
expenditures.

Slope coefficients (102 ) and [test-statistics]

from regressions of the form rtj ¼ b0 xtj þ Etj

Independent (1) (2) (3) (4) (5) (6) (7)

variable

Profits-to-assets 0.85 0.78 0.86 0.68

[5.21] [4.75] [5.17] [3.59]
Accruals  1.35  1.56  1.62  1.79
[  4.67] [  5.43] [  4.21] [  4.61]
R&D-to-market 2.31 2.00 2.00 1.82
[4.22] [3.71] [3.72] [3.44]
log(B/M) 0.36 0.25 0.30 0.29 0.34 0.23 0.26
[5.37] [4.07] [3.85] [4.48] [4.27] [3.09] [3.29]
log(ME)  0.09  0.11  0.11  0.10  0.10  0.12  0.11
[  2.10] [  2.58] [  2.19] [  2.31] [  2.03] [  2.46] [  2.30]
r 1,0  5.09  5.18  5.68  5.29  5.79  5.94  6.03
[  11.0] [  11.1] [  12.1] [  11.4] [  12.5] [  12.7] [  13.1]
r 12,2 0.61 0.60 0.59 0.55 0.54 0.49 0.44
[2.79] [2.77] [2.64] [2.54] [2.45] [2.27] [2.08]

A.4. High frequency strategies
Firms’ revenues, costs of goods sold, and assets are
available on a quarterly basis beginning in 1972 (Compu-
stat data items REVTQ, COGSQ, and ATQ, respectively),
allowing for the construction of gross profitability strate-
gies using more current public information than that
employed when constructing the strategies presented in
Table 2. This section shows that a gross profitability
strategy formed on the basis of the most recently avail-
able public information is even more profitable.
Table A6 shows results of time series regressions
employing a conventional gross profitability strategy,
profitable minus unprofitable (PMU), built following the
Fama and French convention of rebalancing at the end of
June using accounting data from the previous calendar
year, and a high frequency strategy, PMUhf, rebalanced
each month using the most recently released data
(formed on the basis of (REVTQ  COGSQ)/ATQ, from
Compustat quarterly data, employed starting at the end of
the month following a firm’s report date of quarterly
earnings, item RDQ).
The table shows that while the low frequency strategy
generated significant excess returns over the sample
(January 1972 to December 2010, determined by the
availability of the quarterly data), the high frequency
strategy was almost twice as profitable. The high fre-
quency strategy generated excess returns of almost 8% per
year. It also had a larger Fama and French three-factor
alpha and a significant information ratio relative to the
low frequency strategy.
Despite these facts, the remainder of this paper focuses
on gross profitability measured using annual data. I am
particularly interested in the persistent power gross
profitability has predicting returns and its relation to
the similarly persistent value effect. While the high
frequency gross profitability strategy is most profitable
in the months immediately following portfolio formation,
Fig. A1 shows that its profitability persists for more than
three years. Focusing on the strategy formed using annual
profitability data ensures that results are truly driven by
the level of profitability, and not surprises about profit-
ability like those that drive post earnings announcement
drift. The low frequency profitability strategy also incurs
lower transaction costs, turning over only once every four
years, less frequently than the corresponding value strat-
egy, and only a quarter as often as the high frequency
profitability strategy. Using the annual data has the
additional advantage of extending the sample 10 years.
A.5. Profitability and profitability growth
Current profitability, and in particular gross profit-
ability, has power predicting long term growth in gross
profits, earnings, free cash flows and payouts (dividends
plus repurchases), all of which are important determi-
nants of future stock prices. Gross profits-to-assets in
particular is strongly associated with contemporaneous
valuation ratios, so variables that forecast gross profit
growth can be expected to predict future valuations and,
thus, returns.
Table A7 reports results of Fama and MacBeth (1973)
regressions of profitability growth on current profitability.
The table considers both the three- and 10-year growth
rates and employs four different measures of profitability:
gross profits, earnings before extraordinary items, free
cash flow, and total payouts to equity holders (dividends
plus share repurchases). Gross profits, which is an asset-
level measure of profitability, is scaled by assets, while
the other three measures (earnings, free cash flows, and
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28 21

Table A6 West standard errors, with two or nine lags. The data are
Spanning tests employing the high frequency gross profitability strategy. annual and cover 1962–2010.
Results of time series regressions employing low and high frequency
gross profitable-minus-unprofitable strategies. The low frequency strat-
egy, PMU, is rebalanced at the end of each June using the annual A.6. Double sorts on profitability and book-to-market split
Compustat data for the fiscal year ending in the previous calendar year. by size
The high frequency strategy, PMUhf, is rebalanced each month using the
most recent quarterly Compustat data, which are assumed to be Table 6 shows that profitability strategies constructed
available at the end of the month following a firm’s report date of
quarterly earnings. Both strategies are long/short extreme deciles from a
within book-to-market quintiles are more profitable than
sort on gross profitability, using NYSE breaks. Portfolio returns are the unconditional profitability strategy, and value strate-
value-weighted. Regressors include the three Fama and French factors gies constructed within profitability quintiles are more
[the market factor (MKT), the size factor small-minus-large (SMB), and profitable than the unconditional value strategy. The
the value factor high-minus-low (HML)]. Test-statistics are provided in
book-to-market sort yields a great deal of variation in
square brackets. The sample covers January 1972 to December 2010,
with the start date determined by the availability of the quarterly data. firm size, however, especially among the more profitable
stocks, making the results more difficult to interpret. The
PMUhf as PMU as next two tables address this by double sorting on profit-
dependent variable dependent variable ability and book-to-market within the large and small cap
Independent (1) (2) (3) (4) (5) (6)
universes, respectively, where these are defined as firms
variable with market capitalizations above and below the NYSE
median. The gross profits-to-assets and book-to-market
Intercept 0.63 0.78 0.42 0.33 0.51  0.03 breaks are determined using all large or small non-
[3.73] [4.61] [3.10] [2.06] [3.24] [  0.23] financial stocks (NYSE, Amex, and Nasdaq).
MKT  0.14  0.05
Table A8 shows the large cap results, which are largely
[  3.72] [  1.40]
SMB  0.14  0.20 consistent with the all-stock results presented in Table 6.
[  2.63] [  3.93] Again, controlling for profitability improves the perfor-
HML  0.12  0.27 mance of value strategies and controlling for book-to-
[  2.12] [  5.07]
market improves the performance of profitability strate-
PMU 0.64
[16.3]
gies. The average large cap value spread across gross
PMUhf 0.57 profits-to-assets quintiles is 0.59% per month and, in
[16.3] every book-to-market quintile, exceeds the 0.30% per
Adj. R2 4.7% 36.2% 6.7% 36.2% month spread generated by the unconditional large cap
value strategy. The average large cap profitability spread
across book-to-market quintiles is 0.45% per month and,
14
in every book-to-market quintile, exceeds the 0.29% per
Average cumulative excess returns (percent)

12 month spread generated by the unconditional large cap

10
8
6 consequently very thin.
4 somewhat from the all-stock results presented in Table 6.
2
0 to-market has little impact on the performance of profit-
12 24 36
Months from portfolio formation
Fig. A1. Persistence of profitability strategy performance. This figure
shows the average cumulative returns to the high frequency gross
profitability strategy considered in Table A6 from one to 60 months
after portfolio formation. The sample excludes financial firms, and
covers January 1972 to December 2010.
payouts) are scaled by book equity. Regressions include
controls for valuations and size [ln(B/M) and ln(ME)], as
well as prior year’s stock performance. The sample
excludes financial firms (those with one-digit SIC codes
of six). To avoid undue influence from outlying observa-
tions, I trim the independent variables at the 1% and 99%
levels. To avoid undue influence from small firms, I
exclude firms with market capitalizations under $25
million. Test-statistics are calculated using Newey and
profitability strategy. These results should be treated
cautiously, however, as among large cap stocks there are
very few profitable value firms (though there are plenty of
unprofitable growth firms), and the high-high corner is
Table A9 shows the small cap results, which differ
Here controlling for profitability has little impact on the
performance of value strategies, and controlling for book-
48 60 ability strategies. The average small cap value spread
across gross profits-to-assets quintiles is 0.81% per month,
only slightly higher than the 0.72% per month spread
generated by the unconditional small cap value strategy.
The average small cap profitability spread across book-to-
market quintiles is 0.54% per month, slightly less than the
0.57% per month spread generated by the unconditional
small cap profitability strategy. The value effect is stron-
ger, however, among unprofitable stocks, and the profit-
ability effect is stronger among growth stocks.
A.7. Factors constructed controlling for industries
Table 1 suggests that industry-adjusted gross profit-
ability has more power than gross profitability predicting
the cross section of expected returns. This fact suggests
that strategies formed on the basis of industry-adjusted
 22
Table A7
Profitability and profitability growth.
This table reports results of Fama and MacBeth regressions of three- and 10-year growth in profitability, measured by gross profits-to-assets (GP/A), and earnings before extraordinary items (IB), free cash flow
(FCF¼ NIþDP  WCAPCH CAPX), and dividends plus share repurchases (DIV ¼DVC þPRSTKCC) all scaled by book equity (BE), on current profitability, size [ln(ME)], valuations [ln(B/M)] and prior year’s stock
price performance (r 12,1 ). Independent variables are trimmed at the 1% and 99% levels. Test-statistics are calculated using Newey and West standard errors, with two or nine lags. The sample excludes financial
firms (those with one-digit standard industrial classification codes of six) and firms with market capitalizations smaller than $25 million, and uses accounting data for fiscal years ending between 1962 and 2010,
inclusive.

Slope coefficients and [test-statistics] from regressions of the form yt ¼ b0 xtj þ Etj

Regressions predicting three-year growth Regressions predicting 10-year growth

Independent (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
variable

GPt þ t GPt

R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Panel A: Regressions predicting gross profit growth, yt ¼
At
GP/A 0.31 0.33 1.60 1.80
[13.3] [13.6] [6.40] [8.08]
IB/BE 0.10 0.06 0.48 0.15
[2.17] [1.38] [1.18] [0.36]
FCF/BE  0.01  0.05  0.10  0.14
[  0.73] [  3.94] [  0.54] [  0.91]
DIV/BE  0.55  0.81  4.70  5.80
[  4.88] [  5.69] [  9.32] [  8.62]
ln(B/M)  0.05  0.06  0.06  0.05  0.04  0.33  0.36  0.36  0.33  0.26
[  10.7] [  10.8] [  10.4] [  9.83] [  10.9] [  8.00] [  10.4] [  10.1] [  10.6] [  6.64]
ln(ME)  0.12  0.15  0.16  0.16  0.12  0.79  0.96  0.99  1.00  0.79
[  11.2] [  10.9] [  10.7] [  9.96] [  11.8] [  9.61] [  11.8] [  15.8] [  16.6] [  8.99]
r 12,1 0.03 0.04 0.05 0.05 0.03  0.01 0.02 0.02  0.00 0.02
[3.91] [2.71] [2.93] [3.04] [3.36] [  0.07] [0.25] [0.26] [  0.00] [0.26]

IBt þ t IBt
Panel B: Regressions predicting earnings growth, yt ¼
BEt
GP/A 0.04 0.16 0.60 0.45
[1.23] [4.10] [1.57] [1.87]
IB/BE  0.29  0.35 0.63 0.56
[  4.75] [  4.51] [1.37] [1.56]
FCF/BE  0.11 0.03 0.16  0.00
[  2.45] [1.04] [0.66] [  0.02]
DIV/BE  0.26  0.00  0.21  0.36
[  2.66] [  0.04] [  0.31] [  0.56]
ln(B/M)  0.02 0.00  0.00  0.01 0.00 0.07 0.05 0.06 0.06 0.06
[  2.46] [0.22] [  0.55] [  1.99] [0.31] [0.69] [0.62] [0.72] [0.68] [0.82]
ln(ME)  0.04 0.01 0.01  0.04 0.01 0.02  0.02  0.01  0.03 0.04
[  2.54] [0.29] [0.39] [  2.52] [0.54] [0.10] [  0.14] [  0.06] [  0.20] [0.23]
r 12,1  0.09  0.01  0.02  0.10 0.00  0.61  0.66  0.58  0.59  0.65
[  3.56] [  0.32] [  1.31] [  3.94] [0.09] [  1.25] [  1.20] [  1.18] [  1.24] [  1.19]
 FCFt þ t FCFt
Panel C: Regressions predicting free cash flow growth, yt ¼
BEt
GP/A 0.14 0.30 1.10 0.85
[2.78] [6.6] [2.78] [4.14]
IB/BE  0.14 0.08 1.20 1.20
[  1.28] [0.95] [2.08] [2.21]
FCF/BE  0.36  0.50  0.12  0.78
[  5.49] [  6.82] [  0.45] [  3.24]
DIV/BE 0.31 0.77 2.10 2.00
[1.55] [3.85] [2.74] [2.13]
ln(B/M) 0.03 0.04 0.05 0.03 0.04 0.25 0.19 0.21 0.22 0.19
[3.28] [5.1] [6.33] [3.41] [5.52] [2.47] [2.44] [2.17] [2.20] [2.41]
ln(ME) 0.04 0.07 0.09 0.04 0.11 0.44 0.29 0.29 0.37 0.42
[1.35] [3.49] [5.11] [1.45] [5.81] [2.28] [1.61] [1.46] [1.90] [2.05]
r 12,1  0.13  0.05  0.03  0.12  0.04  0.60  0.74  0.59  0.53  0.71
[  4.73] [  3.26] [  1.84] [  4.59] [  2.96] [  1.03] [  1.10] [  1.03] [  0.93] [  1.07]

R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

DIVt þ t DIVt
Panel D: Regressions predicting payout growth, yt ¼
BEt
GP/A 0.01 0.02 0.08 0.09
[2.11] [2.73] [2.77] [2.20]
IB/BE 0.04 0.06 0.11 0.09
[2.54] [3.20] [1.27] [0.65]
FCF/BE 0.01 0.01  0.02 0.02
[2.78] [1.97] [  0.94] [0.36]
DIV/BE  0.28  0.36  0.44  0.60
[  5.53] [  7.41] [  2.72] [  3.28]
ln(B/M)  0.00  0.00  0.00 0.00 0.00  0.02  0.02  0.01  0.01  0.01
[  0.23] [  0.26] [  0.29] [1.95] [2.56] [  3.53] [  4.68] [  4.49] [  2.52] [  1.11]
ln(ME)  0.01  0.01  0.01  0.02  0.01  0.13  0.13  0.13  0.15  0.11
[  3.47] [  4.36] [  5.76] [  7.00] [  5.76] [  8.21] [  6.41] [  6.96] [  8.62] [  5.49]
r 12,1 0.01 0.01 0.01 0.01 0.00 0.00 0.01 0.01  0.00 0.01
[3.64] [3.91] [4.82] [3.50] [1.90] [0.18] [0.48] [1.62] [  0.07] [1.65]

23
24 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Table A8
Double sorts on gross profits-to-assets and book-to-market, large stocks.
This table shows the value-weighted average excess returns to large cap portfolios double sorted on gross profits-to-assets and book-to-market, as well
as results of time series regressions of both sorts’ high minus low portfolios’ returns on the Fama and French factors [the market, size and value factors
MKT, SMB (small-minus-large), and HML (high-minus-low)]. Test statistics are given in square brackets. Large cap is defined as bigger than the NYSE
median. The table also shows the average number of firms, and the average size of firms, in each portfolio (the portfolios exhibit little gross-profits to
asset variation within book-to-market quintiles and little book-to-market variation within profitability quintiles). The sample excludes financial firms
(those with one-digit standard industrial classification codes of six) and covers 1963–2010.

Panel A: Portfolio average returns and time series regression results

Gross profits-to-asset quintiles Profitability strategies

e
Low 2 3 4 High r a bmkt bsmb bhml

Book-to-market quintiles
Low 0.12 0.23 0.29 0.28 0.55 0.43 0.59  0.27  0.14 0.00
[2.21] [3.11] [  6.14] [  2.33] [0.00]
2 0.23 0.31 0.41 0.69 0.87 0.64 0.65  0.14 0.2  0.00
[3.53] [3.52] [  3.16] [3.39] [  0.05]
3 0.38 0.39 0.74 0.62 0.76 0.38 0.19 0.07 0.42 0.10
[1.94] [0.99] [1.59] [6.79] [1.56]
4 0.47 0.58 0.91 1.07 0.81 0.34 0.25 0.08 0.39  0.14
[1.51] [1.15] [1.54] [5.47] [  1.81]
High 0.61 0.78 0.89 1.07 1.04 0.48 0.48  0.03 0.20  0.13
[1.70] [1.69] [  0.41] [2.11] [  1.24]
Value strategies
re 0.49 0.55 0.60 0.79 0.51
[2.19] [2.72] [2.99] [3.47] [1.83]
a 0.23 0.19 0.34 0.40 0.10
[1.21] [1.05] [1.73] [1.91] [0.38]
bmkt  0.22  0.04  0.02  0.06 0.03
[  4.98] [  1.02] [  0.45] [  1.17] [0.44]
bsmb 0.01 0.22 0.24 0.63 0.35
[0.17] [3.78] [3.76] [9.14] [4.04]
bhml 0.89 0.79 0.52 0.61 0.76
[13.5] [12.4] [7.52] [8.13] [8.02]

Panel B: Portfolio average number of firms and average firm size (millions of dollars)
Gross profits-to-asset quintiles Gross profits-to-asset quintiles

Low 2 3 4 High Low 2 3 4 High

Number of firms Average firm size

B/M quintiles
Low 32 34 47 78 140 3,145 3,313 3,958 5,542 5,611
2 33 40 52 60 49 3,377 3,529 3,450 3,114 2,117
3 42 43 43 32 22 2,570 3,094 3,320 2,101 1,264
4 56 44 26 14 8 2,340 3,045 2,666 1,737 1,578
High 38 26 14 8 4 2,338 2,108 2,697 1,256 1,057

characteristics should outperform similar strategies con-
structed on the basis of unadjusted characteristics. If this
is true, then the industry-adjusted strategies might
explain the performance of conventional strategies, in
the sense that the conventional strategies might not
generate abnormal returns relative to the industry-
adjusted strategies, while the conventional strategies
have no hope of explaining the performance of the
industry-adjusted strategies.
Cohen and Polk (1998), Asness, Porter, and Stevens
(2000), and Novy-Marx (2009, 2011) all consider strate-
gies formed on the basis of industry-adjusted book-to-
market. Asness, Porter, and Stevens (2000) also consider
strategies formed on industry-adjusted past performance.
These papers find that strategies formed on the basis of
industry-adjusted book-to-market and past performance
outperform their conventional counterparts. These
industry-adjusted strategies do not, however, generate
higher average returns. Their improved performance is
driven by a reduction in the strategies’ volatilities. While
this is undeniably an important determinant of perfor-
mance, it raises questions regarding whether the
industry-adjusted characteristics are more strongly asso-
ciated with expected returns. Strategies formed on the
basis of industry-adjusted characteristics are much more
balanced across industries. The improved performance of
industry-adjusted value and momentum strategies could
come simply from reducing the strategies’ exposure to
industry related-volatility unrelated to average returns.
While I consider strategies formed on the basis of
industry-adjusted characteristics, I also consider an alter-
native adjustment for industry exposure. This alternative
adjustment simply involves hedging away the industry
exposure from strategies formed on the basis of conven-
tional characteristics. That is, these strategies are formed
by assigning stocks to the portfolios on the basis of
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28 25

Table A9
Double sorts on gross profits-to-assets and book-to-market, small stocks.
This table shows the value-weighted average excess returns to small cap portfolios double sorted on gross profits-to-assets and book-to-market, as
well as results of time series regressions of both sorts’ high minus low portfolios’ returns on the Fama and French factors [the market, size and value
factors MKT, SMB (small-minus-large), and HML (high-minus-low)]. Test statistics are given in square brackets. Small cap is defined as smaller than the
NYSE median. The table also shows the average number of firms, and the average size of firms, in each portfolio (the portfolios exhibit little gross-profits
to asset variation within book-to-market quintiles and little book-to-market variation within profitability quintiles). The sample excludes financial firms
(those with one-digit standard industrial classification codes of six) and covers 1963 to 2010.

Panel A: Portfolio average returns and time series regression results

Gross profits-to-asset quintiles Profitability strategies

e
Low 2 3 4 High r a bmkt bsmb bhml

Book-to-market quintiles
Low  0.20  0.10 0.28 0.37 0.74 0.94 0.98  0.02  0.24 0.08
[4.82] [5.01] [  0.52] [  3.68] [1.11]
2 0.41 0.67 0.67 0.75 1.02 0.61 0.50 0.07  0.17 0.30
[2.79] [2.29] [1.37] [  2.35] [3.90]
3 0.60 0.69 0.72 0.97 1.09 0.49 0.37 0.12 0.08 0.14
[2.99] [2.19] [3.01] [1.42] [2.32]
4 0.75 0.86 0.99 1.07 1.12 0.37 0.21 0.08 0.32 0.11
[2.42] [1.38] [2.29] [6.45] [2.00]
High 0.77 1.00 1.13 1.16 1.08 0.31 0.36  0.14 0.21  0.11
[2.01] [2.32] [  3.80] [4.12] [  1.98]
Value strategies
re 0.98 1.10 0.84 0.79 0.35
[3.95] [4.99] [4.10] [4.57] [1.91]
a 0.68 0.82 0.65 0.60 0.05
[3.73] [4.51] [4.14] [4.44] [0.42]
bmkt  0.02  0.08  0.18  0.13  0.13
[  0.47] [  1.77] [  4.75] [  4.09] [  4.37]
bsmb  0.54  0.22  0.28  0.19  0.09
[  9.00] [  3.73] [  5.48] [  4.26] [  2.21]
bhml 1.13 0.92 0.87 0.74 0.94
[17.5] [14.3] [15.6] [15.5] [20.5]

Panel B: Portfolio average number of firms and average firm size (millions of dollars)
Gross profits-to-asset quintiles Gross profits-to-asset quintiles

Low 2 3 4 High Low 2 3 4 High

Number of firms Average firm size

B/M quintiles
Low 163 67 81 117 203 85 95 108 119 113
2 71 55 78 110 143 89 120 118 116 101
3 71 62 84 112 120 99 122 115 102 85
4 88 85 102 114 110 98 103 89 75 63
High 136 126 121 112 105 68 73 66 57 48

unadjusted characteristics and holding offsetting posi-
tions of equal magnitudes in each stocks’ industry (i.e.,
the Fama and French 49 value-weighted industry portfo-
lios). This helps identify the true importance of industry-
adjusting characteristics, by quantifying the extent to
which performance can be improved by simply reducing
industry driven volatility unrelated to expected returns.
The strategies hedged of industry exposure and the hedge
portfolios also represent a clean decomposition of the
conventional strategies’ returns into intra-industry and
industry components, which makes it simple to quantify
how much of the conventional strategies’ variation is due
to industry exposure.
Table A10 presents the performance of strategies
formed on the basis of unadjusted characteristics, strate-
gies formed on the basis of unadjusted characteristics but
hedged for industry exposure, the previous strategies’
industry-hedges, strategies formed on the basis of
characteristics demeaned by industry, strategies formed
on the basis of the mean industry characteristics, and
strategies formed on the basis of characteristics
demeaned by industry and hedged for industry exposure.
All strategies are formed using the procedure employed in
the construction of HML or UMD. Panel A employs book-
to-market as the primary sorting characteristic. Panel B
employs performance over the first 11 months of the
preceding year. Panel C employs gross profits-to-assets
and, because the strategies are constructed employing
industry adjustments, includes financial firms.
The first column of Table A10 shows the average
excess returns to HML-like factors constructed on the
basis of unadjusted book-to-market, past performance,
and gross profitability. That is, it shows the performance
of the standard Fama and French factors HML and UMD,
as well as a profitable minus unprofitable factor, PMU.
Over the sample, which covers July 1963 to December
26 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28

Table A10
Factors constructed with industry controls.
This table reports the average excess returns (with test-statistics in square brackets) to industry-adjusted factors, constructed employing the HML
(high-minus-low) construction methodology and the results of regressions of the standard factors on these alternative factors’ returns. Panels A, B, and C
show results for strategies formed on the basis of book-to-market, performance over the first 11 months of the preceding year, and gross profits-to-
assets. These strategies are denoted HML (high-minus-low), UMD (up-minus-down), and PMU (profitable-minus-unprofitable), respectively. The first
column presents the standard strategies (i.e., no industry adjustments). The second column shows strategies hedged for industry exposure, in which each
stock position is offset with an opposite position in the firm’s industry (Fama and French 49, value-weighted). The third column shows the industry
hedge. The fourth and fifth columns show strategies constructed using a tertile sort on the primary sorting characteristic demeaned by industry, and
sorted on the industry characteristic, respectively. The sixth column shows strategies constructed by sorting on the characteristic demeaned by industry
and hedged for industry exposure. The sample covers July 1963 to December 2010.

Methodology used in factor construction

Sorted on
industry-adjusted
characteristic
Hedged Sorted on Sorted on and hedged
for industry The industry industry-adjusted industry for industry
Standard exposure hedge characteristic characteristic exposure

Panel A: Alternative HML returns, and results from regressions of HML on these alternatives
E½r e  0.40 0.36 0.04 0.38 0.07 0.42
[3.25] [6.09] [0.65] [5.03] [0.51] [6.67]
Intercept  0.20 0.33  0.03 0.35  0.24
[  2.52] [6.11] [  0.36] [5.14] [  2.95]
Slope 1.64 1.68 1.14 0.75 1.53
[31.1] [49.3] [23.3] [36.5] [29.4]
2
Adj. R 62.9% 81.0% 48.8% 70.0% 60.3%

Panel B: Alternative UMD returns, and results from regressions of UMD on these alternatives
E½r e  0.72 0.62 0.17 0.62 0.54 0.61
[3.94] [5.42] [2.37] [5.17] [3.50] [5.28]
Intercept  0.21 0.34  0.13 0.21  0.19
[  3.17] [3.76] [  1.56] [1.89] [  3.19]
Slope 1.49 2.23 1.36 0.93 1.49
[63.8] [41.8] [48.5] [31.2] [69.8]
Adj. R2 87.7% 75.4% 80.5% 63.0% 89.5%

Panel C: Alternative PMU returns, and results from regressions of PMU on these alternatives
E½r e  0.23 0.15 0.08 0.27 0.12 0.26
[2.41] [3.56] [0.84] [4.42] [1.05] [5.43]
Intercept 0.14 0.16 0.11 0.15 0.03
[1.50] [3.77] [1.14] [2.58] [0.35]
Slope 0.60 0.92 0.47 0.65 0.76
[6.60] [48.9] [7.33] [30.9] [9.86]
Adj. R2 7.0% 80.8% 8.5% 62.6% 14.5%

2010, HML generates average excess returns of 0.40% per
month, with a test-statistic equal to 3.25, and has a
realized annual Sharpe ratio of 0.47. UMU generates
average excess returns of 0.72% per month, with a test-
statistic of 3.94, and has a realized annual Sharpe ratio of
0.57. PMU generates average excess returns of 0.23% per
month, with a test-statistic equal to 2.41, and has a
realized annual Sharpe ratio of 0.35.
The second column shows the performance of the
strategies hedged of industry exposure. Hedging the
strategies decreases the average returns generated by all
three strategies but increases all three strategies’ Sharpe
ratios. While hedged HML, UMD and PMU generate excess
average returns over the sample of only 0.36%, 0.62%, and
0.15% per month, respectively, the strategies’ realized
annual Sharpe ratios are 0.88, 0.79, and 0.52, far in excess
of their conventional counterparts. In all three cases, the
strategies either price or over-price their conventional
counterparts. HML and UMD have significant negative
abnormal returns relative to the hedged strategies, while
PMU has statistically insignificant returns relative to the
hedged strategy.
The third column shows the performance of the
hedges. The results here contrast strongly with those
presented in the second column. Only the momentum
strategy generates significant excess average returns, and
these are relatively modest. That is, while there is some
momentum at the industry level, industry average book-
to-market and industry average profitability appear totally
unrelated to expected returns. Even so, the industry-
related components contribute most of the volatility of
HML and PMU. While contributing only 10% (0.04/0.40) of
HML’s average excess returns, industry exposure drives
80% of the factor’s variation. Similarly, industry exposure
contributes only 34% (0.08/0.23) of PMU’s average excess
returns but, again, drives more than 80% of its variation.
The fourth and fifth columns show the performance of
the strategies constructed on the basis of characteristics
demeaned by industry and industry average characteris-
tics, respectively. Column four shows that sorting on
 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28 27

industry-adjusted characteristics improves the perfor-
mance of all three strategies. For the value and momen-
tum strategies this improvement is slightly less
pronounced, however, than that achieved by simply hed-
ging for industry exposure. This suggests that much of the
benefit realized by forming strategies on the basis of
industry-adjusted book-to-market and past performance
comes simply from reducing the strategies’ industry
exposures. With gross profitability the situation is very
different. Industry-adjusting gross profitability reduces
the volatility of the associated factor while simulta-
neously increasing its average returns, suggesting that
industry-adjusted profitability is truly more strongly
associated with average excess returns. The strategy
formed on the basis of industry adjusted gross profit-
ability generates excess average returns a third higher
than the unadjusted strategy, 0.27% per year with a
test-statistic of 4.42, and has a higher Sharpe ratio
of 0.64.
The sixth column shows that hedging the remaining
industry exposure of the strategies formed on the basis of
the industry-adjusted characteristics further improves the
strategies’ performances. This is especially true for PMU
and, to a lesser extent, HML. The average annual Sharpe
ratios of the strategies formed on the basis of industry-
adjusted book-to-market, past performance, and gross
profitability, and hedged for industry exposure, are 0.97,
0.77, and 0.79, respectively, much higher than the 0.47,
0.57, and 0.35 achieved by their conventional counter-
parts. The performance of these strategies suggests that it
is worthwhile investigating whether they have any power
to explain anomalies.

Table A11
Anomaly strategy abnormal returns relative to alternative models.
This table reports the average excess returns (E½r e ) to strategies formed by sorting on the variables used in factor construction (market capitalization,
book-to-market, performance over the first 11 months of the preceding year, and gross profits-to-assets demeaned by industry and hedged for industry
exposure); earnings related-variables [return on assets, earnings-to-price, asset turnover, gross margins, standardized unexpected earnings (SUE)]; and
failure probability, default risk (Ohlson’s O-score), net stock issuance, asset growth, total accruals, and organizational capital. Strategies are long–short
extreme deciles from a sort on the corresponding variable, employing NYSE breaks, and returns are value-weighted. Momentum, return on assets, return
on equity, SUE, failure, and default probability strategies are rebalanced monthly, while the other strategies are rebalanced annually, at the end of June.
Strategies based on variables scaled by assets exclude financial firms. The table also reports abnormal returns relative to the Fama and French four-factor
model, the Fama and French four-factor model plus the industry-adjusted profitability factor (PMUn), the three-factor model employing the market and
industry-adjusted value and momentum factors (HMLn and UMDn), and the four-factor model employing the market and industry-adjusted value,
momentum, and profitability factors. Test-statistics are provided in square brackets. The sample covers July 1973 through December 2010, and is
determined by the availability of quarterly earnings data.

Abnormal returns relative to the model employing the factors:

Sorting variable used MKT, SMB, MKT, SMB, HML, MKT, HMLn, MKT, HMLn,
in strategy construction E½r e  HML, UMD UMD, PMUn UMDn UMDn, PMUn

Market equity 0.35  0.19  0.02  0.13 0.38

[1.53] [  1.60] [  0.14] [  0.52] [1.54]
Book-to-market 0.58 0.05 0.27  0.20  0.03
[3.13] [0.39] [1.92] [  1.32] [  0.20]
Prior performance 1.43 0.52 0.53  0.11  0.14
[4.28] [3.99] [3.89] [  0.76] [  0.92]
Industry-adjusted 0.21 0.32  0.02 0.34  0.04
profitability [2.34] [4.01] [  0.23] [3.95] [  0.61]
Return on assets 0.67 0.84 0.20 0.73  0.15
[2.81] [4.63] [1.20] [3.19] [  0.75]
Return on equity 1.02 0.82 0.41 0.64 0.07
[4.47] [4.23] [2.13] [2.91] [0.32]
Asset turnover 0.54 0.47  0.32 0.63  0.15
[2.92] [2.45] [  1.96] [3.12] [  0.81]
Gross margins 0.02 0.42 0.16 0.39 0.01
[0.15] [3.33] [1.28] [2.52] [0.05]
SUE 0.69 0.54 0.45 0.48 0.36
[4.00] [3.66] [2.94] [3.03] [2.20]
Failure probability 0.76 0.94 0.40 0.57  0.26
[2.09] [4.44] [1.94] [2.09] [  0.99]
Ohlson’s O-score 0.11 0.59 0.47 0.41 0.09
[0.58] [4.55] [3.47] [2.26] [0.48]
Net stock issuance 0.73 0.62 0.38 0.52 0.21
[5.15] [4.70] [2.86] [3.66] [1.49]
Total accruals 0.37 0.37 0.60 0.24 0.39
[2.35] [2.32] [3.69] [1.41] [2.16]
Asset growth 0.70 0.30 0.36 0.16 0.23
[4.17] [2.12] [2.41] [1.02] [1.37]
Organizational capital 0.46 0.30 0.14 0.36 0.27
[3.73] [2.55] [1.15] [2.75] [1.94]
Root mean squared
pricing error 0.67 0.54 0.36 0.44 0.22
28 R. Novy-Marx / Journal of Financial Economics 108 (2013) 1–28
A.8. Anomaly performance relative to alternative models
Table 10 shows the excess returns to 15 test strategies,
these strategies’ abnormal returns relative to the standard
four-factor Fama and French model, and the alternative
four-factor model employing the market and industry-
adjusted value, momentum, and profitability factors. That
table shows that the root mean squared pricing error
across the 15 anomalies is much lower for the alternative
four-factor model than for the standard four-factor model,
0.22% per month compared with 0.54% per month. This
section investigates how much of this improved perfor-
mance comes from the inclusion of a profitability factor
and how much comes from industry-adjusting the value
and momentum factors.
The second column of Table A11 again shows the 15
strategies’ excess returns. The third column shows the
strategies’ abnormal returns relative to the Fama and
French four-factor model, and the last column shows the
abnormal returns relative to the alternative four-factor
model. The fourth and fifth columns show the perfor-
mance of two intermediate models: the five-factor model
formed by augmenting the standard four-factor model
with the industry-adjusted profitability factor, and the
three-factor model composed of the market and industry-
adjusted value and momentum factors. The root mean
squared pricing errors of the 15 anomaly strategies
relative to these two models are 0.36% and 0.44%, respec-
tively. This suggests that almost two-thirds of the
improved performance of the alternative four-factor
model is due to the inclusion of the profitability factor,
and only roughly one third is due to the industry-
adjustments to the value and momentum factors.
References
Asness, C.S., Porter, R.B., Stevens, R., 2000. Predicting stock returns using
industry-relative firm characteristics. Unpublished working paper.
AQR Capital Management, Greenwich, CT.
Ball, R., 1978. Anomalies in relationships between securities’ yields and
yield-surrogates. Journal of Financial Economics 3, 103–126.
Berk, J.B., 1995. A critique of size-related anomalies. Review of Financial
Studies 8, 275–286.
Campbell, J.Y., Hilscher, J., Szilagyi, J., 2008. In search of distress risk.
Journal of Finance 63, 2899–2939.
Carlson, M., Fisher, A., Giammarino, R., 2004. Corporate investment and
asset pricing dynamics: implications for the cross section of returns.
Journal of Finance 56, 2577–2603.
Chan, K., Chan, L.K.C., Jegadeesh, N., Lakonishok, J., 2006. Earnings quality
and stock returns. Journal of Business 79, 1041–1082.
Chan, L.K.C., Lakonishok, J., Sougiannis, T., 2001. The stock market
valuation of research and development expenditures. Journal of
Finance 56, 2431–2456.
Chen, L., Novy-Marx, R., Zhang, L., 2010. An alternative three–factor
model. Unpublished working paper.
Cohen, R.B., Polk, C.K., 1998. The impact of industry factors in asset-
pricing tests. Unpublished working paper.
Eisfeldt, A.L., Papanikolaou, D., 2011. Organization capital and the cross
section of expected returns. Unpublished working paper.
Fama, E.F., French, K.R., 1993. Common risk factors in the returns on
stocks and bonds. Journal of Financial Economics 33, 3–56.
Fama, E.F., French, K.R., 1997. Industry costs of equity. Journal of
Financial Economics 43, 153–193.
Fama, E.F., French, K.R., 2006. Profitability, investment and average
returns. Journal of Financial Economics 82, 491–518.
Fama, E.F., French, K.R., 2008. Dissecting anomalies. Journal of Finance
63, 1653–1678.
Fama, E.F., MacBeth, J.D., 1973. Risk, return, and equilibrium: empirical
tests. Journal of Political Economy 81, 607–636.
Graham, B., Dodd, D., 1934. Security Analysis. McGraw-Hill, New York.
Lakonishok, J., Shleifer, A., Vishny, R., 1994. Contrarian investment,
extrapolation, and risk. Journal of Finance 49, 1541–1578.
Lettau, M., Wachter, J.A., 2007. Why is long-horizon equity less risky?
A duration-based explanation of the value premium. Journal of
Finance 62, 55–92.
Novy-Marx, R., 2009. Competition, productivity, organization and the
cross section of expected returns. Unpublished working paper.
Novy-Marx, R., 2011. Operating leverage. Review of Finance 15, 103–134.
Ohlson, J.A., 1980. Financial ratios and the probabilistic prediction of
bankruptcy. Journal of Accounting Research 18, 109–131.
Sloan, R.G., 1996. Do stock prices fully reflect information in accruals and
cash flows about future earnings? Accounting Review 71, 289–315.
Zhang, L., 2005. The value premium. Journal of Finance 60, 67–103.